From 37a7eea82b5cd20053d34a7570eb0a80ecd60580 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:41:22 +0000
Subject: [PATCH 01/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/Code?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../Pix2Vox_ID1284_for_TensorFlow/Code/.keep | 0
.../Code/binvox_rw.py | 284 -------
.../Code/demo_AttSets.py | 186 -----
.../Code/export_obj.py | 45 --
.../Code/main_AttSets.py | 719 -----------------
.../Code/main_AttSets_bak.py | 722 ------------------
.../Code/modelarts_entry.py | 55 --
.../Code/npu_train.sh | 59 --
.../Code/ops_info.json | 10 -
.../Code/tools.py | 482 ------------
.../Code/voxel.py | 89 ---
11 files changed, 2651 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/.keep
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/binvox_rw.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/demo_AttSets.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/export_obj.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/main_AttSets.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/main_AttSets_bak.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/modelarts_entry.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/npu_train.sh
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/ops_info.json
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/tools.py
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/voxel.py
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/.keep
deleted file mode 100644
index e69de29bb..000000000
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/binvox_rw.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/binvox_rw.py
deleted file mode 100644
index 40005d24a..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/binvox_rw.py
+++ /dev/null
@@ -1,284 +0,0 @@
-# Copyright (C) 2012 Daniel Maturana
-# This file is part of binvox-rw-py.
-#
-# binvox-rw-py is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# binvox-rw-py is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-#
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-# along with binvox-rw-py. If not, see .
-#
-
-"""
-Binvox to Numpy and back.
->>> import numpy as np
->>> import binvox_rw
->>> with open('chair.binvox', 'rb') as f:
-... m1 = binvox_rw.read_as_3d_array(f)
-...
->>> m1.dims
-[32, 32, 32]
->>> m1.scale
-41.133000000000003
->>> m1.translate
-[0.0, 0.0, 0.0]
->>> with open('chair_out.binvox', 'wb') as f:
-... m1.write(f)
-...
->>> with open('chair_out.binvox', 'rb') as f:
-... m2 = binvox_rw.read_as_3d_array(f)
-...
->>> m1.dims==m2.dims
-True
->>> m1.scale==m2.scale
-True
->>> m1.translate==m2.translate
-True
->>> np.all(m1.data==m2.data)
-True
->>> with open('chair.binvox', 'rb') as f:
-... md = binvox_rw.read_as_3d_array(f)
-...
->>> with open('chair.binvox', 'rb') as f:
-... ms = binvox_rw.read_as_coord_array(f)
-...
->>> data_ds = binvox_rw.dense_to_sparse(md.data)
->>> data_sd = binvox_rw.sparse_to_dense(ms.data, 32)
->>> np.all(data_sd==md.data)
-True
->>> # the ordering of elements returned by numpy.nonzero changes with axis
->>> # ordering, so to compare for equality we first lexically sort the voxels.
->>> np.all(ms.data[:, np.lexsort(ms.data)] == data_ds[:, np.lexsort(data_ds)])
-True
-"""
-from npu_bridge.npu_init import *
-
-import numpy as np
-
-class Voxels(object):
- """ Holds a binvox model.
- data is either a three-dimensional numpy boolean array (dense representation)
- or a two-dimensional numpy float array (coordinate representation).
- dims, translate and scale are the model metadata.
- dims are the voxel dimensions, e.g. [32, 32, 32] for a 32x32x32 model.
- scale and translate relate the voxels to the original model coordinates.
- To translate voxel coordinates i, j, k to original coordinates x, y, z:
- x_n = (i+.5)/dims[0]
- y_n = (j+.5)/dims[1]
- z_n = (k+.5)/dims[2]
- x = scale*x_n + translate[0]
- y = scale*y_n + translate[1]
- z = scale*z_n + translate[2]
- """
-
- def __init__(self, data, dims, translate, scale, axis_order):
- self.data = data
- self.dims = dims
- self.translate = translate
- self.scale = scale
- assert (axis_order in ('xzy', 'xyz'))
- self.axis_order = axis_order
-
- def clone(self):
- data = self.data.copy()
- dims = self.dims[:]
- translate = self.translate[:]
- return Voxels(data, dims, translate, self.scale, self.axis_order)
-
- def write(self, fp):
- write(self, fp)
-
-def read_header(fp):
- """ Read binvox header. Mostly meant for internal use.
- """
- line = fp.readline().strip()
- if not line.startswith(b'#binvox'):
- raise IOError('Not a binvox file')
- dims = list(map(int, fp.readline().strip().split(b' ')[1:]))
- translate = list(map(float, fp.readline().strip().split(b' ')[1:]))
- scale = list(map(float, fp.readline().strip().split(b' ')[1:]))[0]
- line = fp.readline()
- return dims, translate, scale
-
-def read_as_3d_array(fp, fix_coords=True):
- """ Read binary binvox format as array.
- Returns the model with accompanying metadata.
- Voxels are stored in a three-dimensional numpy array, which is simple and
- direct, but may use a lot of memory for large models. (Storage requirements
- are 8*(d^3) bytes, where d is the dimensions of the binvox model. Numpy
- boolean arrays use a byte per element).
- Doesn't do any checks on input except for the '#binvox' line.
- """
- dims, translate, scale = read_header(fp)
- raw_data = np.frombuffer(fp.read(), dtype=np.uint8)
- # if just using reshape() on the raw data:
- # indexing the array as array[i,j,k], the indices map into the
- # coords as:
- # i -> x
- # j -> z
- # k -> y
- # if fix_coords is true, then data is rearranged so that
- # mapping is
- # i -> x
- # j -> y
- # k -> z
- values, counts = raw_data[::2], raw_data[1::2]
- data = np.repeat(values, counts).astype(np.bool)
- data = data.reshape(dims)
- if fix_coords:
- # xzy to xyz TODO the right thing
- data = np.transpose(data, (0, 2, 1))
- axis_order = 'xyz'
- else:
- axis_order = 'xzy'
- return Voxels(data, dims, translate, scale, axis_order)
-
-def read_as_coord_array(fp, fix_coords=True):
- """ Read binary binvox format as coordinates.
- Returns binvox model with voxels in a "coordinate" representation, i.e. an
- 3 x N array where N is the number of nonzero voxels. Each column
- corresponds to a nonzero voxel and the 3 rows are the (x, z, y) coordinates
- of the voxel. (The odd ordering is due to the way binvox format lays out
- data). Note that coordinates refer to the binvox voxels, without any
- scaling or translation.
- Use this to save memory if your model is very sparse (mostly empty).
- Doesn't do any checks on input except for the '#binvox' line.
- """
- dims, translate, scale = read_header(fp)
- raw_data = np.frombuffer(fp.read(), dtype=np.uint8)
-
- values, counts = raw_data[::2], raw_data[1::2]
-
- sz = np.prod(dims)
- index, end_index = 0, 0
- end_indices = np.cumsum(counts)
- indices = np.concatenate(([0], end_indices[:-1])).astype(end_indices.dtype)
-
- values = values.astype(np.bool)
- indices = indices[values]
- end_indices = end_indices[values]
-
- nz_voxels = []
- for index, end_index in zip(indices, end_indices):
- nz_voxels.extend(range(index, end_index))
- nz_voxels = np.array(nz_voxels)
- # TODO are these dims correct?
- # according to docs,
- # index = x * wxh + z * width + y; // wxh = width * height = d * d
-
- x = nz_voxels / (dims[0]*dims[1])
- zwpy = nz_voxels % (dims[0]*dims[1]) # z*w + y
- z = zwpy / dims[0]
- y = zwpy % dims[0]
- if fix_coords:
- data = np.vstack((x, y, z))
- axis_order = 'xyz'
- else:
- data = np.vstack((x, z, y))
- axis_order = 'xzy'
-
- #return Voxels(data, dims, translate, scale, axis_order)
- return Voxels(np.ascontiguousarray(data), dims, translate, scale, axis_order)
-
-def dense_to_sparse(voxel_data, dtype=np.int):
- """ From dense representation to sparse (coordinate) representation.
- No coordinate reordering.
- """
- if voxel_data.ndim!=3:
- raise ValueError('voxel_data is wrong shape; should be 3D array.')
- return np.asarray(np.nonzero(voxel_data), dtype)
-
-def sparse_to_dense(voxel_data, dims, dtype=np.bool):
- if voxel_data.ndim!=2 or voxel_data.shape[0]!=3:
- raise ValueError('voxel_data is wrong shape; should be 3xN array.')
- if np.isscalar(dims):
- dims = [dims]*3
- dims = np.atleast_2d(dims).T
- # truncate to integers
- xyz = voxel_data.astype(np.int)
- # discard voxels that fall outside dims
- valid_ix = ~np.any((xyz < 0) | (xyz >= dims), 0)
- xyz = xyz[:,valid_ix]
- out = np.zeros(dims.flatten(), dtype=dtype)
- out[tuple(xyz)] = True
- return out
-
-#def get_linear_index(x, y, z, dims):
- #""" Assuming xzy order. (y increasing fastest.
- #TODO ensure this is right when dims are not all same
- #"""
- #return x*(dims[1]*dims[2]) + z*dims[1] + y
-
-def write(voxel_model, fp):
- """ Write binary binvox format.
- Note that when saving a model in sparse (coordinate) format, it is first
- converted to dense format.
- Doesn't check if the model is 'sane'.
- """
- if voxel_model.data.ndim==2:
- # TODO avoid conversion to dense
- dense_voxel_data = sparse_to_dense(voxel_model.data, voxel_model.dims)
- else:
- dense_voxel_data = voxel_model.data
-
- fp.write('#binvox 1\n')
- fp.write('dim '+' '.join(map(str, voxel_model.dims))+'\n')
- fp.write('translate '+' '.join(map(str, voxel_model.translate))+'\n')
- fp.write('scale '+str(voxel_model.scale)+'\n')
- fp.write('data\n')
- if not voxel_model.axis_order in ('xzy', 'xyz'):
- raise ValueError('Unsupported voxel model axis order')
-
- if voxel_model.axis_order=='xzy':
- voxels_flat = dense_voxel_data.flatten()
- elif voxel_model.axis_order=='xyz':
- voxels_flat = np.transpose(dense_voxel_data, (0, 2, 1)).flatten()
-
- # keep a sort of state machine for writing run length encoding
- state = voxels_flat[0]
- ctr = 0
- for c in voxels_flat:
- if c==state:
- ctr += 1
- # if ctr hits max, dump
- if ctr==255:
- fp.write(chr(state))
- fp.write(chr(ctr))
- ctr = 0
- else:
- # if switch state, dump
- fp.write(chr(state))
- fp.write(chr(ctr))
- state = c
- ctr = 1
- # flush out remainders
- if ctr > 0:
- fp.write(chr(state))
- fp.write(chr(ctr))
-
-if __name__ == '__main__':
- import doctest
- doctest.testmod()
-
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/demo_AttSets.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/demo_AttSets.py
deleted file mode 100644
index aad6f0277..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/demo_AttSets.py
+++ /dev/null
@@ -1,186 +0,0 @@
-#
-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-from npu_bridge.npu_init import *
-import tensorflow as tf
-import os
-import sys
-sys.path.append('..')
-import tools as tools
-import numpy as np
-import matplotlib.pyplot as plt
-
-GPU='0'
-
-
-def load_real_rgbs(test_mv=5):
- obj_rgbs_folder ='/home/ma-user/work/3DOR/Data/amazon_real_rgbs/airfilter/'
- rgbs = []
- rgbs_views = sorted(os.listdir(obj_rgbs_folder))
- for v in rgbs_views:
- if not v.endswith('png'): continue
-
- rgbs.append(tools.Data.load_single_X_rgb_r2n2(obj_rgbs_folder + v, train=False))
-
- rgbs = np.asarray(rgbs)
- x_sample = rgbs[0:test_mv, :, :, :].reshape(1, test_mv, 127, 127, 3)
- return x_sample, None
-
-def load_shapenet_rgbs(test_mv=8):
- obj_rgbs_folder = './Data_sample/ShapeNetRendering/03001627/1a6f615e8b1b5ae4dbbc9440457e303e/rendering/'
- obj_gt_vox_path ='./Data_sample/ShapeNetVox32/03001627/1a6f615e8b1b5ae4dbbc9440457e303e/model.binvox'
- rgbs=[]
- rgbs_views = sorted(os.listdir(obj_rgbs_folder))
- for v in rgbs_views:
- if not v.endswith('png'): continue
- rgbs.append(tools.Data.load_single_X_rgb_r2n2(obj_rgbs_folder + v, train=False))
- rgbs = np.asarray(rgbs)
- x_sample = rgbs[0:test_mv, :, :, :].reshape(1, test_mv, 127, 127, 3)
- y_true = tools.Data.load_single_Y_vox(obj_gt_vox_path)
- return x_sample, y_true
-
-def ttest_demo():
-# model_path = './Model_released/'
- model_path='/home/ma-user/work/3DOR/Model/train_mod/'
- if not os.path.isfile(model_path + 'model.cptk.data-00000-of-00001'):
- print ('please download our released model first!')
- return
-
- config = tf.ConfigProto(allow_soft_placement=True)
- config.gpu_options.visible_device_list = GPU
-
-
-
- with tf.Session(config=npu_config_proto(config_proto=config)) as sess:
- saver = tf.train.import_meta_graph(model_path + 'model.cptk.meta', clear_devices=True)
- saver.restore(sess, model_path + 'model.cptk')
- print ('model restored!')
-
- # graph = tf.get_default_graph()
- # print(graph.get_operations())
-
- X = tf.get_default_graph().get_tensor_by_name("Placeholder:0")
- Y_pred = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_9:0")
-
- plot_data_8 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_8:0")
- plot_data_7 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_7:0") #############(1,1024)
- plot_data_6 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_6:0") #############(1,1024)
- plot_data_5 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_5:0")
- plot_data_4 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_4:0")
- plot_data_3 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_3:0")
- plot_data_2 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_2:0")
- plot_data_1 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_1:0")
-
-
-# print("X: ", X.shape) #Tensor("Placeholder:0", shape=(?, ?, 127, 127, 3), dtype=float32)
-# print(Y_pred) #Tensor("r2n/Reshape_9:0", shape=(?, 32, 32, 32), dtype=float32)
-
-# print("x_sample: ", x_sample.shape)
-# print("x_sample_data: ", type(x_sample[:,:,:,:,1]))
-# print(y_pred.shape) ###############################(1, 32, 32, 32) ##############################
-
-
-# x_sample, gt_vox = load_shapenet_rgbs()
- x_sample, gt_vox = load_real_rgbs()
-
-
- plot_buf_1= tf.reshape(plot_data_1, [-1, 32, 32, 1])
- plot_buf_2= tf.reshape(plot_data_2, [-1, 32, 32, 1])
- plot_buf_3= tf.reshape(plot_data_3, [-1, 32, 32, 1])
- plot_buf_4= tf.reshape(plot_data_4, [-1, 32, 32, 1])
- plot_buf_5= tf.reshape(plot_data_5, [-1, 32, 32, 1])
- plot_buf_6= tf.reshape(plot_data_6, [-1, 32, 32, 1])
- plot_buf_7= tf.reshape(plot_data_7, [-1, 32, 32, 1])
- plot_buf_8= tf.reshape(plot_data_8, [-1, 32, 32, 1])
-
-
-# tf.summary.image("RESHAPE_1", plot_buf_1)
-# tf.summary.image("RESHAPE_2", plot_buf_2)
-# tf.summary.image("RESHAPE_3", plot_buf_3)
-# tf.summary.image("RESHAPE_4", plot_buf_4)
-# tf.summary.image("RESHAPE_5", plot_buf_5)
-# tf.summary.image("RESHAPE_6", plot_buf_6)
-# tf.summary.image("RESHAPE_7", plot_buf_7)
-# tf.summary.image("RESHAPE_8", plot_buf_8)
-
-
- summary_8 = tf.summary.image("RESHAPE_8", plot_buf_8)
- summary_7 = tf.summary.image("RESHAPE_7", plot_buf_7)
- summary_6 = tf.summary.image("RESHAPE_6", plot_buf_6)
- summary_5 = tf.summary.image("RESHAPE_5", plot_buf_5)
- summary_4 = tf.summary.image("RESHAPE_4", plot_buf_4)
- summary_3 = tf.summary.image("RESHAPE_3", plot_buf_3)
- summary_2 = tf.summary.image("RESHAPE_2", plot_buf_2)
- summary_1 = tf.summary.image("RESHAPE_1", plot_buf_1)
-
-
-# summary_op = tf.summary.image("RESHAPE_4", plot_buf_4)
-# with tf.Session() as sess:
-# y_pred,1_summary,2_summary = sess.run([Y_pred,summary_op_1,summary_op_2], feed_dict={X: x_sample})
-
- y_pred,summary_pred_1,summary_pred_2,summary_pred_3,summary_pred_4,summary_pred_5,summary_pred_6,summary_pred_7,summary_pred_8 = sess.run([Y_pred,summary_1,summary_2,summary_3,summary_4,summary_5,summary_6,summary_7,summary_8], feed_dict={X: x_sample})
-
-# Write summary tf.summary.FileWriter
- writer = tf.summary.FileWriter('./logs')
-
-
- writer.add_summary(summary_pred_1)
- writer.add_summary(summary_pred_2)
- writer.add_summary(summary_pred_3)
- writer.add_summary(summary_pred_4)
- writer.add_summary(summary_pred_5)
- writer.add_summary(summary_pred_6)
- writer.add_summary(summary_pred_7)
- writer.add_summary(summary_pred_8)
-
- writer.close()
-
-# sys.exit(). sys.exit()
- ###### to visualize
- th = 0.25
- y_pred[y_pred>=th]=1
- y_pred[y_pred= 0.5] = 1
- prediction[prediction < 0.5] = 0
- intersection = np.sum(np.logical_and(prediction,gt))
- union = np.sum(np.logical_or(prediction,gt))
- IoU = float(intersection) / float(union)
- return IoU
-#####################################
-def refiner_network(volumes_in):
- with tf.device('/cpu:0'):
- with tf.variable_scope('ref_enc'):
-
- input_volumes_32 = tf.reshape(volumes_in, [-1, vox_res32, vox_res32, vox_res32, 1],name="ref_net_in")
-
- print("input_volumes_32_shape" , input_volumes_32.shape) #input_volumes_32_shape (?,32,32,32,1)
-
- rn1=Conv3D(filters=32, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_c1')(input_volumes_32)
- rn2=BatchNormalization()(rn1)
- rn3=LeakyReLU(alpha=.2)(rn2)
- print("rn3.shape",rn3.shape) # rn3.shape (?, 32, 32, 32, 32)
- volumes_16_l =MaxPooling3D(pool_size=(2, 2, 2),name='ref_m1')(rn3)
-
- print("volumes_16_l_shape" , volumes_16_l.shape) #volumes_16_l_shape (?,16,16,16,32)
-
- rn5=Conv3D(filters=64, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_c2')(volumes_16_l)
- rn6=BatchNormalization()(rn5)
- rn7=LeakyReLU(alpha=.2)(rn6)
- print("rn7.shape",rn7.shape) #rn7.shape (?, 16, 16, 16, 64)
- volumes_8_l =MaxPooling3D(pool_size=(2, 2, 2),name='ref_m2')(rn7)
-
- print("volumes_8_l_shape" ,volumes_8_l.shape) #volumes_8_l_shape (?,8,8,8,64)
-
- rn9=Conv3D(filters=128, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_c3')(volumes_8_l)
- rn10=BatchNormalization()(rn9)
- rn11=LeakyReLU(alpha=.2)(rn10)
- print("rn11.shape",rn11.shape) #rn11.shape (?, 8, 8, 8, 128)
- volumes_4_l =MaxPooling3D(pool_size=(2, 2, 2),name='ref_m3')(rn11)
-
- print("volumes_4_l_shape" , volumes_4_l.shape) #volumes_4_l_shape (?,4,4,4,128)
-
- flatten_features=tf.reshape(volumes_4_l , [-1,8192],name="ref_fc1_in")
- with tf.variable_scope('ref_fc'):
-
- fc1=Dense(units=2048, activation='relu',name='ref_fc1')(flatten_features)
-# fc1=tanh(fc1)
-
- fc1=relu(fc1, alpha=0.0, max_value=None, threshold=0.0)
-
- print("fc1_shape",fc1.shape) #fc1_shape (?,4,4,4,2048)
-
- fc2=Dense(units=8192, activation='relu',name='ref_fc2')(fc1)
-# fc2=tanh(fc2)
- fc2=relu(fc2, alpha=0.0, max_value=None, threshold=0.0)
-
- print("fc2_shape",fc2.shape) #fc2_shape (?,4,4,4,8192)
-
- fc2=tf.reshape(fc2, [-1, 4,4,4,128],name="ref_fc2_out")
-
- with tf.variable_scope('ref_Dec'):
-
- reshaped_1=Add()([fc2,volumes_4_l])
-
- print("reshaped_1.shape",reshaped_1.shape) #reshaped_1.shape (?,4,4,4,128)
-
- rn13=Conv3DTranspose(filters=64, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_d1',strides=(2, 2, 2))(reshaped_1)
-
- rn14=BatchNormalization()(rn13)
- volumes_4_r=relu(rn14, alpha=0.0, max_value=None, threshold=0.0)
-
- print("volumes_4_r_shape",volumes_4_r.shape) #volumes_4_r_shape (?,8,8,8,64)
-
- reshaped_2=Add() ([volumes_4_r,volumes_8_l])
-
- print("reshaped_2_shape",reshaped_2.shape) #volumes_2_shape (?,8,8,8,64)
-
-
- rn16=Conv3DTranspose(filters=32, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_d2',strides=(2, 2, 2))(reshaped_2)
- rn17=BatchNormalization()(rn16)
- volumes_8_r =relu(rn17, alpha=0.0, max_value=None, threshold=0.0)
-
- reshaped_3=Add()([volumes_8_r,volumes_16_l])
-
- print("reshaped_3_shape",reshaped_3.shape) #reshaped_3_shape (?,16,16,16,32)
-
-
- rn19=Conv3DTranspose(filters=1, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_d3',strides=(2, 2, 2))(volumes_8_r)
- print("rn19_shape",rn19.shape) #rn19_shape (?, ?, ?, ?, 1)
-# volumes_16_r= tf.nn.sigmoid(rn19,name='ref_sigmoid1')
-# reshape_4=volumes_16_r ####################
-
- reshape_4=Add()([rn19,input_volumes_32])
- reshape_4=(reshape_4*0.5)
- print("reshape_4_5",reshape_4.shape) #reshape_4_5 (?,32,32,32,1)
-
- reshape_4= tf.nn.sigmoid(reshape_4,name='ref_sigmoid1')
-
-
- print("reshape_4_sig_shape",reshape_4.shape) #reshape_4_sig_shape (?,32,32,32,1)
-
- reshape_5=tf.reshape(reshape_4, [-1, vox_res32, vox_res32, vox_res32],name="ref_out")
-
- return reshape_5
-
-def attsets_fc(x, out_ele_num):
- with tf.variable_scope('att_fc'):
- in_ele_num = tf.shape(x)[1]
- in_ele_len = int(x.get_shape()[2])
- out_ele_len = in_ele_len
- print("out_ele_len ", out_ele_len)
- ####################
- x_1st = x
- x_1st_tp = tf.reshape(x_1st, [-1, in_ele_len],name="att_in")
- weights_1st = tools.Ops.fc(x_1st_tp, out_d=out_ele_num*out_ele_len, name="att")
-
- ########## option 1
- weights_1st = weights_1st
- ########## option 2
-# weights_1st = tf.nn.tanh(weights_1st)
-
- ####################
- weights_1st = tf.reshape(weights_1st, [-1, in_ele_num, out_ele_num, out_ele_len],name="att_fc_out")
- weights_1st = tf.nn.softmax(weights_1st, 1)
- x_1st = tf.tile(x_1st[:,:,None,:], [1,1,out_ele_num,1])
- x_1st = x_1st*weights_1st
- x_1st = tf.reduce_sum(x_1st, axis=1)
- x_1st = tf.reshape(x_1st, [-1, out_ele_num*out_ele_len],name="att_out")
- return x_1st, weights_1st
-
-#####################################
-class Network:
- def __init__(self):
- self.train_mod_dir = './Model/train_mod/'
- self.train_sum_dir = './Model/train_sum/'
- self.test_res_dir = './Model/test_res/'
- self.test_sum_dir = './Model/test_sum/'
-
- print ('re_train : ', re_train)
- if os.path.exists(self.test_res_dir):
- if re_train:
- print ('test_res_dir and files kept!')
- else:
- shutil.rmtree(self.test_res_dir)
- os.makedirs(self.test_res_dir)
- print ('test_res_dir: deleted and then created!')
- else:
- os.makedirs(self.test_res_dir)
- print ('test_res_dir: created!')
-
- if os.path.exists(self.train_mod_dir):
- if re_train:
- if os.path.exists(self.train_mod_dir + 'model.cptk.data-00000-of-00001'):
- print ('model found! will be reused!')
- else:
- print ('model not found! error!')
- #exit()
- else:
- shutil.rmtree(self.train_mod_dir)
- os.makedirs(self.train_mod_dir)
- print ('train_mod_dir: deleted and then created!')
- else:
- os.makedirs(self.train_mod_dir)
- print ('train_mod_dir: created!')
-
- if os.path.exists(self.train_sum_dir):
- if re_train:
- print ('train_sum_dir and files kept!')
- else:
- shutil.rmtree(self.train_sum_dir)
- os.makedirs(self.train_sum_dir)
- print ('train_sum_dir: deleted and then created!')
- else:
- os.makedirs(self.train_sum_dir)
- print ('train_sum_dir: created!')
-
- if os.path.exists(self.test_sum_dir):
- if re_train:
- print ('test_sum_dir and files kept!')
- else:
- shutil.rmtree(self.test_sum_dir)
- os.makedirs(self.test_sum_dir)
- print ('test_sum_dir: deleted and then created!')
- else:
- os.makedirs(self.test_sum_dir)
- print ('test_sum_dir: created!')
-
- def base_r2n2(self, X_rgb):
- with tf.variable_scope('Encoder'):
- im_num = tf.shape(X_rgb)[1]
-
- [_, _, d1, d2, cc] = X_rgb.get_shape()
- X_rgb = tf.reshape(X_rgb, [-1, int(d1), int(d2), int(cc)],name="en_in")
- print("Network Structure")
- print("base_r2n2",X_rgb.shape) #base_r2n2 (?, 127, 127, 3)
-
-# plot_buf_1= tf.reshape(X_rgb, [-1, 127, 127, 3]) ###################
-# tf.summary.image("Input", plot_buf_1)###############################
-
- en_c = [96, 128, 256, 256, 256, 256]
- l1 = tools.Ops.xxlu(tools.Ops.conv2d(X_rgb, k=7, out_c=en_c[0], str=1, name='en_c1'), label='lrelu')
- print("l1_r2n",l1.shape) #l1_r2n (?, 127, 127, 96)
-
-
-
- l2 = tools.Ops.xxlu(tools.Ops.conv2d(l1, k=3, out_c=en_c[0], str=1, name='en_c2'), label='lrelu')
- l2 = tools.Ops.maxpool2d(l2, k=2, s=2, name='en_mp1')
- print("l2_r2n",l2.shape) #l2_r2n (?, 64, 64, 96)
-
-# plot_buf_1= tf.reshape(l2, [-1, 32, 32, 3]) #########################
-# tf.summary.image("L2_MP_en", plot_buf_1)#############################
-
- l3 = tools.Ops.xxlu(tools.Ops.conv2d(l2, k=3, out_c=en_c[1], str=1, name='en_c3'), label='lrelu')
- print("l3_r2n",l3.shape) #l3_r2n (?, 64, 64, 128)
- l4 = tools.Ops.xxlu(tools.Ops.conv2d(l3, k=3, out_c=en_c[1], str=1, name='en_c4'), label='lrelu')
- print("l4_r2n",l4.shape) #l4_r2n (?, 64, 64, 128)
- l22 = tools.Ops.conv2d(l2, k=1, out_c=en_c[1], str=1, name='en_c22')
- print("l22_r2n",l22.shape) #l22_r2n (?, 64, 64, 128)
- l4 = l4 + l22
- l4 = tools.Ops.maxpool2d(l4, k=2, s=2, name='en_mp2')
- print("l4+l22_r2n",l4.shape) #l4+l22_r2n (?, 32, 32, 128)
-
-# plot_buf_1= tf.reshape(l4, [-1, 32, 32, 3]) ##########################
-# tf.summary.image("l4+l22_en", plot_buf_1)#############################
-
- l5 = tools.Ops.xxlu(tools.Ops.conv2d(l4, k=3, out_c=en_c[2], str=1, name='en_c5'), label='lrelu')
- print("l5_r2n",l5.shape) #l5_r2n (?, 32, 32, 256)
- l6 = tools.Ops.xxlu(tools.Ops.conv2d(l5, k=3, out_c=en_c[2], str=1, name='en_c6'), label='lrelu')
- print("l6_r2n",l6.shape) #l6_r2n (?, 32, 32, 256)
- l44 = tools.Ops.conv2d(l4, k=1, out_c=en_c[2], str=1, name='en_c44')
- print("l44_r2n",l44.shape) #l44_r2n (?, 32, 32, 256)
- l6 = l6 + l44
- l6 = tools.Ops.maxpool2d(l6, k=2, s=2, name='en_mp3')
- print("l6+l44_r2n",l6.shape) #l6+l44_r2n (?, 16, 16, 256)
-
-# plot_buf_1= tf.reshape(l6, [-1, 16, 16, 3]) ##########################
-# tf.summary.image("l6+l44_en", plot_buf_1)#############################
-
- l7 = tools.Ops.xxlu(tools.Ops.conv2d(l6, k=3, out_c=en_c[3], str=1, name='en_c7'), label='lrelu')
- print("l7_r2n",l7.shape) #l7_r2n (?, 16, 16, 256)
- l8 = tools.Ops.xxlu(tools.Ops.conv2d(l7, k=3, out_c=en_c[3], str=1, name='en_c8'), label='lrelu')
- print("l8_r2n",l8.shape)#l8_r2n (?, 16, 16, 256)
-# l66=tools.Ops.conv2d(l6, k=1, out_c=en_c[3], str=1, name='en_c66')
-# print("l66_r2n",l66.shape)
-# l8=l8+l66
- l8 = tools.Ops.maxpool2d(l8, k=2, s=2, name='en_mp4')
- print("l8_r2n",l8.shape) #l8_r2n (?, 8, 8, 256)
-
-# plot_buf_1= tf.reshape(l8, [-1, 8, 8, 3]) ########################
-# tf.summary.image("l8_en", plot_buf_1)#############################
-
- l9 = tools.Ops.xxlu(tools.Ops.conv2d(l8, k=3, out_c=en_c[4], str=1, name='en_c9'), label='lrelu')
- print("l9_r2n",l9.shape) #l9_r2n (?, 8, 8, 256)
- l10 = tools.Ops.xxlu(tools.Ops.conv2d(l9, k=3, out_c=en_c[4], str=1, name='en_c10'), label='lrelu')
- print("l10_r2n",l10.shape)#l10_r2n (?, 8, 8, 256)
- l88 = tools.Ops.conv2d(l8, k=1, out_c=en_c[4], str=1, name='en_c88')
- print("l88_r2n",l88.shape) #l88_r2n (?, 8, 8, 256)
- l10 = l10 + l88
- l10 = tools.Ops.maxpool2d(l10, k=2, s=2, name='en_mp5')
- print("l10_r2n",l10.shape) #l10_r2n (?, 4, 4, 256)
-
-# plot_buf_1= tf.reshape(l10, [-1, 4, 4, 3]) ########################
-# tf.summary.image("l10_en", plot_buf_1)#############################
-
- l11 = tools.Ops.xxlu(tools.Ops.conv2d(l10, k=3, out_c=en_c[5], str=1, name='en_c11'), label='lrelu')
- print("l11_r2n",l11.shape) #l11_r2n (?, 4, 4, 256)
- l12 = tools.Ops.xxlu(tools.Ops.conv2d(l11, k=3, out_c=en_c[5], str=1, name='en_c12'), label='lrelu')
- print("l12_r2n",l12.shape) #l12_r2n (?, 4, 4, 256)
- l1010 = tools.Ops.conv2d(l10, k=1, out_c=en_c[5], str=1, name='en_c1010')
- print("l1010_r2n",l1010.shape) #l1010_r2n (?, 4, 4, 256)
- l12 = l12 + l1010
- l12 = tools.Ops.maxpool2d(l12, k=2, s=2, name='en_mp6')
- print("l12_r2n",l12.shape) #l12_r2n (?, 2, 2, 256)
-
-# plot_buf_1= tf.reshape(l12, [-1, 2, 2, 3]) ########################
-# tf.summary.image("l12_en", plot_buf_1)#############################
-
- [_, d1, d2, cc] = l12.get_shape()
- l12 = tf.reshape(l12, [-1, int(d1) * int(d2) * int(cc)],name="en_fc1_in")
- print("fc1_input_r2n",l12.shape) #fc1_input_r2n (?, 1024)
- fc = tools.Ops.xxlu(tools.Ops.fc(l12, out_d=1024, name='en_fc1'), label='lrelu')
- print("fc1_output_r2n",fc.shape)#fc1_output_r2n (?, 1024)
-
- with tf.variable_scope('Att_Net'):
- #### use fc attention
- input = tf.reshape(fc, [-1, im_num, 1024],name="Att_fc_in")
- print("att_fc_in_r2n",input.shape) #att_fc_in_r2n (?, ?, 1024)
- latent_3d, weights = attsets_fc(input, out_ele_num=1)
- print("att_fc_out_r2n",latent_3d.shape) #att_fc_out_r2n (?, 1024)
-
- with tf.variable_scope('Decoder'):
- ####
- latent_3d = tools.Ops.xxlu(tools.Ops.fc(latent_3d, out_d=4*4*4*128, name='de_fc2'), label='lrelu')
- print("fc3_out_r2n",latent_3d.shape) #fc3_out_r2n (?, 8192)
- latent_3d = tf.reshape(latent_3d, [-1, 4, 4, 4, 128],name="de_fc2_out")
-
- ####
-
- de_c = [128, 128, 128, 64, 32, 1]
-
- print("d1_in_r2n",latent_3d.shape) #d1_in_r2n (?, 4, 4, 4, 128)
- d1 = tools.Ops.xxlu(tools.Ops.deconv3d(latent_3d, k=3, out_c=de_c[1], str=2, name='de_c1'), label='lrelu')
- print("d1_out_r2n",d1.shape) #d1_out_r2n (?, 8, 8, 8, 128)
- d2 = tools.Ops.xxlu(tools.Ops.deconv3d(d1, k=3, out_c=de_c[1], str=1, name='de_c2'), label='lrelu')
- print("d2_out_r2n",d2.shape) #d2_out_r2n (?, 8, 8, 8, 128)
- d00 = tools.Ops.deconv3d(latent_3d, k=1, out_c=de_c[1], str=2, name='de_c00')
- print("d00_out_r2n",d00.shape)#d00_out_r2n (?, 8, 8, 8, 128)
- d2 = d2 + d00
- print("d2+d00_out_r2n",d2.shape)#d2+d00_out_r2n (?, 8, 8, 8, 128)
-
-# plot_buf_1= tf.reshape(d2, [-1, 8, 8, 4]) ################################
-# tf.summary.image("d2+d00_out_de", plot_buf_1)#############################
-
- d3 = tools.Ops.xxlu(tools.Ops.deconv3d(d2, k=3, out_c=de_c[2], str=2, name='de_c3'), label='lrelu')
- print("d3_out_r2n",d3.shape)#d3_out_r2n (?, 16, 16, 16, 128)
- d4 = tools.Ops.xxlu(tools.Ops.deconv3d(d3, k=3, out_c=de_c[2], str=1, name='de_c4'), label='lrelu')
- print("d4_out_r2n",d4.shape)#d4_out_r2n (?, 16, 16, 16, 128)
- d22 = tools.Ops.deconv3d(d2, k=1, out_c=de_c[2], str=2, name='de_c22')
- print("d22_out_r2n",d22.shape)#d22_out_r2n (?, 16, 16, 16, 128)
- d4 = d4 + d22
- print("d4+d22_out_r2n",d4.shape)#d4+d22_out_r2n (?, 16, 16, 16, 128)
-
-# plot_buf_1= tf.reshape(d4, [-1, 16, 16, 4]) ##############################
-# tf.summary.image("d4+d22_out_de", plot_buf_1)#############################
-
- d5 = tools.Ops.xxlu(tools.Ops.deconv3d(d4, k=3, out_c=de_c[3], str=2, name='de_c5'), label='lrelu')
- print("d5_out_r2n",d5.shape)#d5_out_r2n (?, 32, 32, 32, 64)
- d6 = tools.Ops.xxlu(tools.Ops.deconv3d(d5, k=3, out_c=de_c[3], str=1, name='de_c6'), label='lrelu')
- print("d6_out_r2n",d6.shape)#d6_out_r2n (?, 32, 32, 32, 64)
- d44 = tools.Ops.deconv3d(d4, k=1, out_c=de_c[3], str=2, name='de_c44')
- print("d44_out_r2n",d44.shape)#d44_out_r2n (?, 32, 32, 32, 64)
- d6 = d6 + d44
- print("d6+d44_out_r2n",d6.shape) #d6+d44_out_r2n (?, 32, 32, 32, 64)
-
-# plot_buf_1= tf.reshape(d6, [-1, 32, 32, 4]) ##############################
-# tf.summary.image("d6+d44_out_de", plot_buf_1)#############################
-
- d7 = tools.Ops.xxlu(tools.Ops.deconv3d(d6, k=3, out_c=de_c[4], str=1, name='de_c7'), label='lrelu')
- print("d7_out_r2n",d7.shape) #d7_out_r2n (?, 32, 32, 32, 32)
- d8 = tools.Ops.xxlu(tools.Ops.deconv3d(d7, k=3, out_c=de_c[4], str=1, name='de_c8'), label='lrelu')
- print("d8_out_r2n",d8.shape)#d8_out_r2n (?, 32, 32, 32, 32)
- d77 = tools.Ops.xxlu(tools.Ops.deconv3d(d7, k=3, out_c=de_c[4], str=1, name='de_c77'), label='lrelu')
- print("d77_out_r2n",d77.shape)#d77_out_r2n (?, 32, 32, 32, 32)
- d8 = d8 + d77
- print("d8+d77_out_r2n",d8.shape) #d8+d77_out_r2n (?, 32, 32, 32, 32)
-
-# plot_buf_1= tf.reshape(d8, [-1, 32, 32, 4]) ##############################
-# tf.summary.image("d8+d77_out_de", plot_buf_1)#############################
-
- d11 = tools.Ops.deconv3d(d8, k=3, out_c=de_c[5], str=1, name='de_c9')
- print("d11_out_r2n",d11.shape) #d11_out_r2n (?, 32, 32, 32, 1)
-
-# plot_buf_1= tf.reshape(d11, [-1, 32, 32,4]) ##########################
-# tf.summary.image("Ref_input", plot_buf_1)#############################
-
- ref_in = tf.reshape(d11, [-1, vox_res32, vox_res32, vox_res32],name="ref_in") ###
-
- y = tf.nn.sigmoid(d11,name='de_sigmoid')
-
- att_o = tf.reshape(y, [-1, vox_res32, vox_res32, vox_res32],name="de_out")
-
- print("att_out_shape",att_o.shape) #att_out_shape (?, 32, 32, 32)
-
- with tf.variable_scope('ref_net'):
-
- ref_o=refiner_network(ref_in)
-
-# plot_buf_1= tf.reshape(ref_o, [-1, 32, 32,4]) ######################
-# tf.summary.image("Ref_Out", plot_buf_1)#############################
-
- return ref_o,att_o, weights
-
- def build_graph(self):
- img_res = 127
- vox_res = 32
- self.X_rgb = tf.placeholder(shape=[None, None, img_res, img_res, 3], dtype=tf.float32)
- self.Y_vox = tf.placeholder(shape=[None, vox_res, vox_res, vox_res], dtype=tf.float32)
- self.lr = tf.placeholder(tf.float32)
- self.refine_lr = tf.placeholder(tf.float32)
- with tf.device('/cpu:0'):
- self.Y_pred,self.vae_o, self.weights = self.base_r2n2(self.X_rgb)
- tf.summary.histogram('Attsets_Weights', self.weights)
- with tf.device('/cpu:0'):
- ### rec loss
- print ("reached")
- with tf.variable_scope('Loss_Fun'):
- Y_vox_ = tf.reshape(self.Y_vox, shape=[-1, vox_res ** 3])
- Y_pred_ = tf.reshape(self.Y_pred, shape=[-1, vox_res ** 3])
- vae_o_=tf.reshape(self.vae_o, shape=[-1, vox_res ** 3])
-
- self.vae_loss = tf.reduce_mean(-tf.reduce_mean(Y_vox_ * tf.log(vae_o_ + 1e-8), reduction_indices=[1]) -
- tf.reduce_mean((1 - Y_vox_) * tf.log(1 - vae_o_ + 1e-8),reduction_indices=[1]))
- self.rec_loss = tf.reduce_mean(-tf.reduce_mean(Y_vox_ * tf.log(Y_pred_ + 1e-8), reduction_indices=[1]) -
- tf.reduce_mean((1 - Y_vox_) * tf.log(1 - Y_pred_ + 1e-8),reduction_indices=[1]))
- sum_rec_loss = tf.summary.scalar('rec_loss', self.rec_loss)
- self.sum_merged = sum_rec_loss
- tf.summary.histogram('rec_loss', self.rec_loss)
- tf.summary.scalar("vae-loss",self.vae_loss)
- tf.summary.histogram("vae_loss",self.vae_loss)
- self.mean_loss = tf.div(x=tf.math.add(x=self.vae_loss,y=self.rec_loss,name='add_loss'),y=2,name='mean_loss')
- tf.summary.histogram("mean_vae_loss",self.mean_loss)
- tf.summary.scalar("mean_vae_loss",self.mean_loss)
-
- with tf.variable_scope('Evaluation_Metric'):
- gt_vox=Y_vox_
- self.iou_ref = metric_iou(Y_pred_,gt_vox)
- tf.summary.scalar('iou_refiner', self.iou_ref)
- tf.summary.histogram('iou_refiner', self.iou_ref)
- self.iou_vae = metric_iou(vae_o_,gt_vox)
- tf.summary.scalar('iou_vae', self.iou_vae)
- tf.summary.histogram("iou_vae",self.iou_vae)
-
- with tf.variable_scope('Optimization'):
-
- base_en_var = [var for var in tf.trainable_variables() if var.name.startswith('Encoder/en')]
- base_dec_var = [var for var in tf.trainable_variables() if var.name.startswith('Decoder/de')]
- att_var = [var for var in tf.trainable_variables() if var.name.startswith('Att_Net/att')]
- refine_var = [var for var in tf.trainable_variables() if var.name.startswith('ref_net/ref')]
- self.refine_optim = tf.train.AdamOptimizer(learning_rate=self.refine_lr).minimize(self.rec_loss, var_list=refine_var)
- self.base_en_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=base_en_var)
- self.base_de_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=base_dec_var)
- self.att_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=att_var)
-
-
- print ("total weights:",tools.Ops.variable_count())
- self.saver = tf.train.Saver(max_to_keep=1)
-
- config = tf.ConfigProto(allow_soft_placement=True)
- #config.gpu_options.allow_growth = True
- #config.gpu_options.visible_device_list = '0,1'
-
- self.sess = tf.Session(config=npu_config_proto(config_proto=config))
- self.merged = tf.summary.merge_all()
- self.sum_writer_train = tf.summary.FileWriter(self.train_sum_dir, self.sess.graph)
- self.sum_writer_test = tf.summary.FileWriter(self.test_sum_dir, self.sess.graph)
-
- #######################
- path = self.train_mod_dir
- #path = './Model_released/' # retrain the released model
-
- if os.path.isfile(path + 'model.cptk.data-00000-of-00001'):
- print ("restoring saved model!")
- self.saver.restore(self.sess, path + 'model.cptk')
- else:
- self.sess.run(tf.global_variables_initializer())
-
-
- return 0
-
- def train(self, data):
- for epoch in range(0, obs_config.epoch, 1):
- train_view_num = obs_config.train_view_num
- data.shuffle_train_files(epoch, train_mv=train_view_num)
- total_train_batch_num = data.total_train_batch_num #int(len(self.X_rgb_train_files)/(self.batch_size*train_mv))
- print ('total_train_batch_num:', total_train_batch_num)
- for i in range(total_train_batch_num):
- #### training
- X_rgb_bat, Y_vox_bat = data.load_X_Y_train_next_batch(train_mv=train_view_num)
- print("multi_view_train_X_rgb_bat : ",X_rgb_bat.shape)#np.asarray(X.append(X_rgb[b*train_mv:(b+1)*train_mv,:]))
-
-
- print(time.ctime())
-
- ##### option 1: seperate train, seperate optimize
- #if epoch<=30:
- # single_view_train=True
- # multi_view_train=False
- #else:
- # single_view_train=False
- # multi_view_train=True
-
- ##### optiion 2: joint train, seperate optimize
- single_view_train = True
- multi_view_train = False
-
- if epoch <= 5:
- att_lr=.0002
- ref_lr=.0002
- if epoch > 5 and epoch <= 40:
- att_lr=.0001
- ref_lr=.0001
- if epoch > 40:
- att_lr=.00005
- ref_lr=.00005
- ########### single view train
- if single_view_train:
-
- rgb = np.reshape(X_rgb_bat,[batch_size*train_view_num, 1, 127,127,3])
- print("single_view_train_rgb_input_shape ",rgb.shape)
- vox = np.tile(Y_vox_bat[:,None,:,:,:],[1,train_view_num,1,1,1])
- vox = np.reshape(vox, [batch_size*train_view_num, 32,32,32])
- start_time=time.time()
- vae_loss_c,eee,ddd, rec_loss_c, sum_train,rrr,mean_vae,iou_ref_,iou_vae_ = self.sess.run([self.vae_loss,self.base_en_optim2,self.base_de_optim2,self.rec_loss,self.merged,self.refine_optim,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr,self.refine_lr: ref_lr})
- print('TimeHistory = %.4f' % (time.time()-start_time))
- print ('ep:', epoch, 'i:', i, 'train single rec loss:', rec_loss_c)
- print ('ep:', epoch, 'i:', i, 'train single vae loss:', vae_loss_c)
- print ('ep:', epoch, 'i:', i, 'train single mean_vae loss:',mean_vae)
- print ('ep:', epoch, 'i:', i, 'train single ref_iou:',iou_ref_)
- print ('ep:', epoch, 'i:', i, 'train single vae_iou:',iou_vae_)
-
- ########## multi view train
- if multi_view_train:
-
- vae_loss_c,rec_loss_c, _, sum_train,xxx,mean_vae,iou_ref_,iou_vae_ = self.sess.run([self.vae_loss,self.rec_loss, self.att_optim2, self.merged,self.refine_optim,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: X_rgb_bat, self.Y_vox: Y_vox_bat,self.lr: att_lr,self.refine_lr: ref_lr})
- print ('ep:', epoch, 'i:', i, 'train multi rec loss:', rec_loss_c)
- print ('ep:', epoch, 'i:', i, 'train multi vae loss:', vae_loss_c)
- print('ep:',epoch,'i',i,'train multi mean_vae loss:',mean_vae)
- print ('ep:', epoch, 'i:', i, 'train multi ref_iou:',iou_ref_)
- print ('ep:', epoch, 'i:', i, 'train multi vae_iou:',iou_vae_)
-
- ############
- if i % 5 == 0:
- self.sum_writer_train.add_summary(sum_train, epoch * total_train_batch_num + i)
-
-
- #### testing
- if i % 10 == 0 :
- X_rgb_batch, Y_vox_batch = data.load_X_Y_test_next_batch(test_mv=3)
-
-# vae_pred = tf.get_default_graph().get_tensor_by_name("Decoder/de_out:0")
-# ref_pred = tf.get_default_graph().get_tensor_by_name("ref_net/ref_Dec/ref_out:0")
-# gt_vox=Y_vox_batch.astype(np.float32)
-
-# iou_pred = metric_iou(ref_pred,gt_vox)
-# tf.summary.scalar("iou",iou_pred)
-
- rrrr,aaaa,rec_loss_te, qwerty, Y_vox_test_pred, att_pred, sum_test,mean_vae,iou_ref_,iou_vae_ = \
- self.sess.run([self.refine_optim,self.att_optim2,self.rec_loss,self.vae_loss, self.Y_pred,self.weights, self.merged,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: X_rgb_batch, self.Y_vox: Y_vox_batch,self.lr: 0,self.refine_lr: 0})
-
- X_rgb_batch = X_rgb_batch.astype(np.float16)
-
- Y_vox_batch = Y_vox_batch.astype(np.float16)
- Y_vox_test_pred = Y_vox_test_pred.astype(np.float16)
- att_pred = att_pred.astype(np.float16)
- to_save = {'X_test':X_rgb_batch,'Y_test_pred':Y_vox_test_pred,'att_pred':att_pred,'Y_test_true':Y_vox_batch}
- scipy.io.savemat(self.test_res_dir+'X_Y_pred_'+str(epoch).zfill(2)+'_'+str(i).zfill(5)+'.mat',to_save,do_compression=True)
-
-
- self.sum_writer_test.add_summary(sum_test, epoch * total_train_batch_num + i)
-
-# iou_ref=evaluate_voxel_prediction(ref_pred_,gt_vox)
-# iou_vae=evaluate_voxel_prediction(vae_pred_,gt_vox)
-
-# print("Ref_iou:",iou_ref)
-# print("Vae_iou:",iou_vae)
-
-# plot_list_iou.append(iou_ref)
-# plot_list_i.append((i/50))
-# graph_plot(plot_list_iou,plot_list_i)
- print ('ep:', epoch, 'i:', i, 'test rec loss:', rec_loss_te)
- print ('ep:', epoch, 'i:', i, 'test vae loss:', qwerty )
- print ('ep:', epoch, 'i:', i, 'test mean_vae loss:', mean_vae)
- print ('ep:', epoch, 'i:', i, 'test ref_iou:',iou_ref_)
- print ('ep:', epoch, 'i:', i, 'test vae_iou:',iou_vae_)
-
- #### model saving
- #if i % 100 == 0 :
- #self.saver.save( self.sess, save_path=self.train_mod_dir + 'model.cptk' )
- #print ( 'epoch:', epoch, 'i:', i, 'model saved!' )
-
-
-# plt.show()
-
-
-##########
-if __name__ =='__main__':
- npu_keras_sess = set_keras_session_npu_config()
-
- net = Network() #net=object to create instance
-
- print("network compleated") ###
-
- net.build_graph()
- print("graph compleated")
-
-
- data = tools.Data(config)
- print("tools.data compleated")
-
-
- print('training data')
-
- net.train(data)
-
- close_session(npu_keras_sess)
-
-
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/main_AttSets_bak.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/main_AttSets_bak.py
deleted file mode 100644
index 65135e0c4..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/main_AttSets_bak.py
+++ /dev/null
@@ -1,722 +0,0 @@
-#
-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-from npu_bridge.npu_init import *
-import tensorflow as tf
-import os
-
-#os.system('pip install scipy==0.19.0')
-#os.system('pip install matplotlib==2.0.2')
-#os.system('pip install skimage==0.13.0')
-#os.system('pip install PyMCubes')
-
-import shutil
-import sys
-import scipy.io
-sys.path.append('..')
-import tools as tools
-import numpy as np
-import time
-from keras.layers import BatchNormalization,Conv3D,MaxPooling3D,Dense,Reshape,Add,LeakyReLU,Conv3DTranspose
-from keras.activations import relu,sigmoid,tanh
-from keras import models
-import copy
-import matplotlib.pyplot as plt
-
-import argparse
-#import moxing as mox
-
-
-parser = argparse.ArgumentParser()
-parser.add_argument('--train_url', type=str, help='the path model saved')
-parser.add_argument("--data_url", type=str, default="./dataset")
-obs_config = parser.parse_args()
-
-###################
-#import tflearn
-###################
-
-batch_size = 2
-img_res = 127
-vox_res32 = 32
-total_mv = 24
-GPU0 = '0'
-GPU1 = '1'
-
-re_train = True
-#re_train=True
-single_view_train = True
-multi_view_train = True
-
-###########################
-
-plot_list_iou = []
-plot_list_i = []
-iii=0
-config={} # python dictionary
-config['batch_size'] = batch_size
-config['total_mv'] = total_mv
-#config['cat_names'] = ['02691156','02828884','04530566','03636649','03001627']
-#config['cat_names'] = ['02691156','02828884','02933112','02958343','03001627','03211117',
-# '03636649','03691459','04090263','04256520','04379243','04401088','04530566']
-config['cat_names'] = ['02691156']
-for name in config['cat_names']:
- config['X_rgb_'+name] = obs_config.data_url+'/ShapeNetRendering/'+name+'/'
-
- config['Y_vox_'+name] = obs_config.data_url+'/ShapeNetVox32/'+name+'/'
-
- # config['Y_vox_'+name] = '/home/wiproec4/3d reconstruction/attsets/Data_sample/shapenet dataset/ShapeNetVox32/train_1_dataset/'+name+'/'
-
-
-# output : {'batch_size': 1, 'total_mv': 24, 'cat_names': ['03001627'], 'Y_vox_03001627': '/home/wiproec4/3d reconstruction/attsets/Data_sample/#ShapeNetVox32/03001627/', 'X_rgb_03001627': '/home/wiproec4/3d reconstruction/attsets/Data_sample/ShapeNetRendering/03001627/'}
-
-def metric_iou(prediction, gt):
-# labels = tf.greater_equal(gt[gt], 0.5)
-# prediction = tf.cast(prediction,tf.int32)
- predictions = tf.greater_equal(prediction, 0.5)
- gt_=tf.greater_equal(gt, 0.5)
- intersection = tf.reduce_sum(tf.cast(tf.logical_and(predictions,gt_),tf.float32))
- union = tf.reduce_sum(tf.cast(tf.math.logical_or(predictions,gt_),tf.float32))
- iou = tf.cast(x = intersection,dtype=tf.float32)/ tf.cast(x = union,dtype=tf.float32)
- return iou
-
-
-def graph_plot(iou_value,i_value):
- x = i_value
- y = iou_value
- plt.plot(x, y, color='green', linestyle='dashed', linewidth = 3, marker='o', markerfacecolor='blue', markersize=12)
-
- plt.ylim(0,2)
- plt.xlim(0,500)
-
- plt.xlabel('Iterations')
-
- plt.ylabel('IOU')
-
- plt.title('Refiner Accuracy')
-
- plt.show()
-
-
-def evaluate_voxel_prediction(prediction, gt):
- #""" The prediction and gt are 3 dim voxels. Each voxel has values 1 or 0"""
- prediction[prediction >= 0.5] = 1
- prediction[prediction < 0.5] = 0
- intersection = np.sum(np.logical_and(prediction,gt))
- union = np.sum(np.logical_or(prediction,gt))
- IoU = float(intersection) / float(union)
- return IoU
-#####################################
-def refiner_network(volumes_in):
- with tf.device('/cpu:0'):
- with tf.variable_scope('ref_enc'):
-
- input_volumes_32 = tf.reshape(volumes_in, [-1, vox_res32, vox_res32, vox_res32, 1],name="ref_net_in")
-
- print("input_volumes_32_shape" , input_volumes_32.shape) #input_volumes_32_shape (?,32,32,32,1)
-
- rn1=Conv3D(filters=32, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_c1')(input_volumes_32)
- rn2=BatchNormalization()(rn1)
- rn3=LeakyReLU(alpha=.2)(rn2)
- print("rn3.shape",rn3.shape) # rn3.shape (?, 32, 32, 32, 32)
- volumes_16_l =MaxPooling3D(pool_size=(2, 2, 2),name='ref_m1')(rn3)
-
- print("volumes_16_l_shape" , volumes_16_l.shape) #volumes_16_l_shape (?,16,16,16,32)
-
- rn5=Conv3D(filters=64, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_c2')(volumes_16_l)
- rn6=BatchNormalization()(rn5)
- rn7=LeakyReLU(alpha=.2)(rn6)
- print("rn7.shape",rn7.shape) #rn7.shape (?, 16, 16, 16, 64)
- volumes_8_l =MaxPooling3D(pool_size=(2, 2, 2),name='ref_m2')(rn7)
-
- print("volumes_8_l_shape" ,volumes_8_l.shape) #volumes_8_l_shape (?,8,8,8,64)
-
- rn9=Conv3D(filters=128, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_c3')(volumes_8_l)
- rn10=BatchNormalization()(rn9)
- rn11=LeakyReLU(alpha=.2)(rn10)
- print("rn11.shape",rn11.shape) #rn11.shape (?, 8, 8, 8, 128)
- volumes_4_l =MaxPooling3D(pool_size=(2, 2, 2),name='ref_m3')(rn11)
-
- print("volumes_4_l_shape" , volumes_4_l.shape) #volumes_4_l_shape (?,4,4,4,128)
-
- flatten_features=tf.reshape(volumes_4_l , [-1,8192],name="ref_fc1_in")
- with tf.variable_scope('ref_fc'):
-
- fc1=Dense(units=2048, activation='relu',name='ref_fc1')(flatten_features)
-# fc1=tanh(fc1)
-
- fc1=relu(fc1, alpha=0.0, max_value=None, threshold=0.0)
-
- print("fc1_shape",fc1.shape) #fc1_shape (?,4,4,4,2048)
-
- fc2=Dense(units=8192, activation='relu',name='ref_fc2')(fc1)
-# fc2=tanh(fc2)
- fc2=relu(fc2, alpha=0.0, max_value=None, threshold=0.0)
-
- print("fc2_shape",fc2.shape) #fc2_shape (?,4,4,4,8192)
-
- fc2=tf.reshape(fc2, [-1, 4,4,4,128],name="ref_fc2_out")
-
- with tf.variable_scope('ref_Dec'):
-
- reshaped_1=Add()([fc2,volumes_4_l])
-
- print("reshaped_1.shape",reshaped_1.shape) #reshaped_1.shape (?,4,4,4,128)
-
- rn13=Conv3DTranspose(filters=64, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_d1',strides=(2, 2, 2))(reshaped_1)
-
- rn14=BatchNormalization()(rn13)
- volumes_4_r=relu(rn14, alpha=0.0, max_value=None, threshold=0.0)
-
- print("volumes_4_r_shape",volumes_4_r.shape) #volumes_4_r_shape (?,8,8,8,64)
-
- reshaped_2=Add() ([volumes_4_r,volumes_8_l])
-
- print("reshaped_2_shape",reshaped_2.shape) #volumes_2_shape (?,8,8,8,64)
-
-
- rn16=Conv3DTranspose(filters=32, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_d2',strides=(2, 2, 2))(reshaped_2)
- rn17=BatchNormalization()(rn16)
- volumes_8_r =relu(rn17, alpha=0.0, max_value=None, threshold=0.0)
-
- reshaped_3=Add()([volumes_8_r,volumes_16_l])
-
- print("reshaped_3_shape",reshaped_3.shape) #reshaped_3_shape (?,16,16,16,32)
-
-
- rn19=Conv3DTranspose(filters=1, kernel_size=(4, 4, 4), padding='same',data_format="channels_last",name='ref_d3',strides=(2, 2, 2))(volumes_8_r)
- print("rn19_shape",rn19.shape) #rn19_shape (?, ?, ?, ?, 1)
-# volumes_16_r= tf.nn.sigmoid(rn19,name='ref_sigmoid1')
-# reshape_4=volumes_16_r ####################
-
- reshape_4=Add()([rn19,input_volumes_32])
- reshape_4=(reshape_4*0.5)
- print("reshape_4_5",reshape_4.shape) #reshape_4_5 (?,32,32,32,1)
-
- reshape_4= tf.nn.sigmoid(reshape_4,name='ref_sigmoid1')
-
-
- print("reshape_4_sig_shape",reshape_4.shape) #reshape_4_sig_shape (?,32,32,32,1)
-
- reshape_5=tf.reshape(reshape_4, [-1, vox_res32, vox_res32, vox_res32],name="ref_out")
-
- return reshape_5
-
-def attsets_fc(x, out_ele_num):
- with tf.variable_scope('att_fc'):
- in_ele_num = tf.shape(x)[1]
- in_ele_len = int(x.get_shape()[2])
- out_ele_len = in_ele_len
- print("out_ele_len ", out_ele_len)
- ####################
- x_1st = x
- x_1st_tp = tf.reshape(x_1st, [-1, in_ele_len],name="att_in")
- weights_1st = tools.Ops.fc(x_1st_tp, out_d=out_ele_num*out_ele_len, name="att")
-
- ########## option 1
- weights_1st = weights_1st
- ########## option 2
-# weights_1st = tf.nn.tanh(weights_1st)
-
- ####################
- weights_1st = tf.reshape(weights_1st, [-1, in_ele_num, out_ele_num, out_ele_len],name="att_fc_out")
- weights_1st = tf.nn.softmax(weights_1st, 1)
- x_1st = tf.tile(x_1st[:,:,None,:], [1,1,out_ele_num,1])
- x_1st = x_1st*weights_1st
- x_1st = tf.reduce_sum(x_1st, axis=1)
- x_1st = tf.reshape(x_1st, [-1, out_ele_num*out_ele_len],name="att_out")
- return x_1st, weights_1st
-
-#####################################
-class Network:
- def __init__(self):
- self.train_mod_dir = './Model/train_mod/'
- self.train_sum_dir = './Model/train_sum/'
- self.test_res_dir = './Model/test_res/'
- self.test_sum_dir = './Model/test_sum/'
-
- print ('re_train : ', re_train)
- if os.path.exists(self.test_res_dir):
- if re_train:
- print ('test_res_dir and files kept!')
- else:
- shutil.rmtree(self.test_res_dir)
- os.makedirs(self.test_res_dir)
- print ('test_res_dir: deleted and then created!')
- else:
- os.makedirs(self.test_res_dir)
- print ('test_res_dir: created!')
-
- if os.path.exists(self.train_mod_dir):
- if re_train:
- if os.path.exists(self.train_mod_dir + 'model.cptk.data-00000-of-00001'):
- print ('model found! will be reused!')
- else:
- print ('model not found! error!')
- #exit()
- else:
- shutil.rmtree(self.train_mod_dir)
- os.makedirs(self.train_mod_dir)
- print ('train_mod_dir: deleted and then created!')
- else:
- os.makedirs(self.train_mod_dir)
- print ('train_mod_dir: created!')
-
- if os.path.exists(self.train_sum_dir):
- if re_train:
- print ('train_sum_dir and files kept!')
- else:
- shutil.rmtree(self.train_sum_dir)
- os.makedirs(self.train_sum_dir)
- print ('train_sum_dir: deleted and then created!')
- else:
- os.makedirs(self.train_sum_dir)
- print ('train_sum_dir: created!')
-
- if os.path.exists(self.test_sum_dir):
- if re_train:
- print ('test_sum_dir and files kept!')
- else:
- shutil.rmtree(self.test_sum_dir)
- os.makedirs(self.test_sum_dir)
- print ('test_sum_dir: deleted and then created!')
- else:
- os.makedirs(self.test_sum_dir)
- print ('test_sum_dir: created!')
-
- def base_r2n2(self, X_rgb):
- with tf.variable_scope('Encoder'):
- im_num = tf.shape(X_rgb)[1]
-
- [_, _, d1, d2, cc] = X_rgb.get_shape()
- X_rgb = tf.reshape(X_rgb, [-1, int(d1), int(d2), int(cc)],name="en_in")
- print("Network Structure")
- print("base_r2n2",X_rgb.shape) #base_r2n2 (?, 127, 127, 3)
-
-# plot_buf_1= tf.reshape(X_rgb, [-1, 127, 127, 3]) ###################
-# tf.summary.image("Input", plot_buf_1)###############################
-
- en_c = [96, 128, 256, 256, 256, 256]
- l1 = tools.Ops.xxlu(tools.Ops.conv2d(X_rgb, k=7, out_c=en_c[0], str=1, name='en_c1'), label='lrelu')
- print("l1_r2n",l1.shape) #l1_r2n (?, 127, 127, 96)
-
-
-
- l2 = tools.Ops.xxlu(tools.Ops.conv2d(l1, k=3, out_c=en_c[0], str=1, name='en_c2'), label='lrelu')
- l2 = tools.Ops.maxpool2d(l2, k=2, s=2, name='en_mp1')
- print("l2_r2n",l2.shape) #l2_r2n (?, 64, 64, 96)
-
-# plot_buf_1= tf.reshape(l2, [-1, 32, 32, 3]) #########################
-# tf.summary.image("L2_MP_en", plot_buf_1)#############################
-
- l3 = tools.Ops.xxlu(tools.Ops.conv2d(l2, k=3, out_c=en_c[1], str=1, name='en_c3'), label='lrelu')
- print("l3_r2n",l3.shape) #l3_r2n (?, 64, 64, 128)
- l4 = tools.Ops.xxlu(tools.Ops.conv2d(l3, k=3, out_c=en_c[1], str=1, name='en_c4'), label='lrelu')
- print("l4_r2n",l4.shape) #l4_r2n (?, 64, 64, 128)
- l22 = tools.Ops.conv2d(l2, k=1, out_c=en_c[1], str=1, name='en_c22')
- print("l22_r2n",l22.shape) #l22_r2n (?, 64, 64, 128)
- l4 = l4 + l22
- l4 = tools.Ops.maxpool2d(l4, k=2, s=2, name='en_mp2')
- print("l4+l22_r2n",l4.shape) #l4+l22_r2n (?, 32, 32, 128)
-
-# plot_buf_1= tf.reshape(l4, [-1, 32, 32, 3]) ##########################
-# tf.summary.image("l4+l22_en", plot_buf_1)#############################
-
- l5 = tools.Ops.xxlu(tools.Ops.conv2d(l4, k=3, out_c=en_c[2], str=1, name='en_c5'), label='lrelu')
- print("l5_r2n",l5.shape) #l5_r2n (?, 32, 32, 256)
- l6 = tools.Ops.xxlu(tools.Ops.conv2d(l5, k=3, out_c=en_c[2], str=1, name='en_c6'), label='lrelu')
- print("l6_r2n",l6.shape) #l6_r2n (?, 32, 32, 256)
- l44 = tools.Ops.conv2d(l4, k=1, out_c=en_c[2], str=1, name='en_c44')
- print("l44_r2n",l44.shape) #l44_r2n (?, 32, 32, 256)
- l6 = l6 + l44
- l6 = tools.Ops.maxpool2d(l6, k=2, s=2, name='en_mp3')
- print("l6+l44_r2n",l6.shape) #l6+l44_r2n (?, 16, 16, 256)
-
-# plot_buf_1= tf.reshape(l6, [-1, 16, 16, 3]) ##########################
-# tf.summary.image("l6+l44_en", plot_buf_1)#############################
-
- l7 = tools.Ops.xxlu(tools.Ops.conv2d(l6, k=3, out_c=en_c[3], str=1, name='en_c7'), label='lrelu')
- print("l7_r2n",l7.shape) #l7_r2n (?, 16, 16, 256)
- l8 = tools.Ops.xxlu(tools.Ops.conv2d(l7, k=3, out_c=en_c[3], str=1, name='en_c8'), label='lrelu')
- print("l8_r2n",l8.shape)#l8_r2n (?, 16, 16, 256)
-# l66=tools.Ops.conv2d(l6, k=1, out_c=en_c[3], str=1, name='en_c66')
-# print("l66_r2n",l66.shape)
-# l8=l8+l66
- l8 = tools.Ops.maxpool2d(l8, k=2, s=2, name='en_mp4')
- print("l8_r2n",l8.shape) #l8_r2n (?, 8, 8, 256)
-
-# plot_buf_1= tf.reshape(l8, [-1, 8, 8, 3]) ########################
-# tf.summary.image("l8_en", plot_buf_1)#############################
-
- l9 = tools.Ops.xxlu(tools.Ops.conv2d(l8, k=3, out_c=en_c[4], str=1, name='en_c9'), label='lrelu')
- print("l9_r2n",l9.shape) #l9_r2n (?, 8, 8, 256)
- l10 = tools.Ops.xxlu(tools.Ops.conv2d(l9, k=3, out_c=en_c[4], str=1, name='en_c10'), label='lrelu')
- print("l10_r2n",l10.shape)#l10_r2n (?, 8, 8, 256)
- l88 = tools.Ops.conv2d(l8, k=1, out_c=en_c[4], str=1, name='en_c88')
- print("l88_r2n",l88.shape) #l88_r2n (?, 8, 8, 256)
- l10 = l10 + l88
- l10 = tools.Ops.maxpool2d(l10, k=2, s=2, name='en_mp5')
- print("l10_r2n",l10.shape) #l10_r2n (?, 4, 4, 256)
-
-# plot_buf_1= tf.reshape(l10, [-1, 4, 4, 3]) ########################
-# tf.summary.image("l10_en", plot_buf_1)#############################
-
- l11 = tools.Ops.xxlu(tools.Ops.conv2d(l10, k=3, out_c=en_c[5], str=1, name='en_c11'), label='lrelu')
- print("l11_r2n",l11.shape) #l11_r2n (?, 4, 4, 256)
- l12 = tools.Ops.xxlu(tools.Ops.conv2d(l11, k=3, out_c=en_c[5], str=1, name='en_c12'), label='lrelu')
- print("l12_r2n",l12.shape) #l12_r2n (?, 4, 4, 256)
- l1010 = tools.Ops.conv2d(l10, k=1, out_c=en_c[5], str=1, name='en_c1010')
- print("l1010_r2n",l1010.shape) #l1010_r2n (?, 4, 4, 256)
- l12 = l12 + l1010
- l12 = tools.Ops.maxpool2d(l12, k=2, s=2, name='en_mp6')
- print("l12_r2n",l12.shape) #l12_r2n (?, 2, 2, 256)
-
-# plot_buf_1= tf.reshape(l12, [-1, 2, 2, 3]) ########################
-# tf.summary.image("l12_en", plot_buf_1)#############################
-
- [_, d1, d2, cc] = l12.get_shape()
- l12 = tf.reshape(l12, [-1, int(d1) * int(d2) * int(cc)],name="en_fc1_in")
- print("fc1_input_r2n",l12.shape) #fc1_input_r2n (?, 1024)
- fc = tools.Ops.xxlu(tools.Ops.fc(l12, out_d=1024, name='en_fc1'), label='lrelu')
- print("fc1_output_r2n",fc.shape)#fc1_output_r2n (?, 1024)
-
- with tf.variable_scope('Att_Net'):
- #### use fc attention
- input = tf.reshape(fc, [-1, im_num, 1024],name="Att_fc_in")
- print("att_fc_in_r2n",input.shape) #att_fc_in_r2n (?, ?, 1024)
- latent_3d, weights = attsets_fc(input, out_ele_num=1)
- print("att_fc_out_r2n",latent_3d.shape) #att_fc_out_r2n (?, 1024)
-
- with tf.variable_scope('Decoder'):
- ####
- latent_3d = tools.Ops.xxlu(tools.Ops.fc(latent_3d, out_d=4*4*4*128, name='de_fc2'), label='lrelu')
- print("fc3_out_r2n",latent_3d.shape) #fc3_out_r2n (?, 8192)
- latent_3d = tf.reshape(latent_3d, [-1, 4, 4, 4, 128],name="de_fc2_out")
-
- ####
-
- de_c = [128, 128, 128, 64, 32, 1]
-
- print("d1_in_r2n",latent_3d.shape) #d1_in_r2n (?, 4, 4, 4, 128)
- d1 = tools.Ops.xxlu(tools.Ops.deconv3d(latent_3d, k=3, out_c=de_c[1], str=2, name='de_c1'), label='lrelu')
- print("d1_out_r2n",d1.shape) #d1_out_r2n (?, 8, 8, 8, 128)
- d2 = tools.Ops.xxlu(tools.Ops.deconv3d(d1, k=3, out_c=de_c[1], str=1, name='de_c2'), label='lrelu')
- print("d2_out_r2n",d2.shape) #d2_out_r2n (?, 8, 8, 8, 128)
- d00 = tools.Ops.deconv3d(latent_3d, k=1, out_c=de_c[1], str=2, name='de_c00')
- print("d00_out_r2n",d00.shape)#d00_out_r2n (?, 8, 8, 8, 128)
- d2 = d2 + d00
- print("d2+d00_out_r2n",d2.shape)#d2+d00_out_r2n (?, 8, 8, 8, 128)
-
-# plot_buf_1= tf.reshape(d2, [-1, 8, 8, 4]) ################################
-# tf.summary.image("d2+d00_out_de", plot_buf_1)#############################
-
- d3 = tools.Ops.xxlu(tools.Ops.deconv3d(d2, k=3, out_c=de_c[2], str=2, name='de_c3'), label='lrelu')
- print("d3_out_r2n",d3.shape)#d3_out_r2n (?, 16, 16, 16, 128)
- d4 = tools.Ops.xxlu(tools.Ops.deconv3d(d3, k=3, out_c=de_c[2], str=1, name='de_c4'), label='lrelu')
- print("d4_out_r2n",d4.shape)#d4_out_r2n (?, 16, 16, 16, 128)
- d22 = tools.Ops.deconv3d(d2, k=1, out_c=de_c[2], str=2, name='de_c22')
- print("d22_out_r2n",d22.shape)#d22_out_r2n (?, 16, 16, 16, 128)
- d4 = d4 + d22
- print("d4+d22_out_r2n",d4.shape)#d4+d22_out_r2n (?, 16, 16, 16, 128)
-
-# plot_buf_1= tf.reshape(d4, [-1, 16, 16, 4]) ##############################
-# tf.summary.image("d4+d22_out_de", plot_buf_1)#############################
-
- d5 = tools.Ops.xxlu(tools.Ops.deconv3d(d4, k=3, out_c=de_c[3], str=2, name='de_c5'), label='lrelu')
- print("d5_out_r2n",d5.shape)#d5_out_r2n (?, 32, 32, 32, 64)
- d6 = tools.Ops.xxlu(tools.Ops.deconv3d(d5, k=3, out_c=de_c[3], str=1, name='de_c6'), label='lrelu')
- print("d6_out_r2n",d6.shape)#d6_out_r2n (?, 32, 32, 32, 64)
- d44 = tools.Ops.deconv3d(d4, k=1, out_c=de_c[3], str=2, name='de_c44')
- print("d44_out_r2n",d44.shape)#d44_out_r2n (?, 32, 32, 32, 64)
- d6 = d6 + d44
- print("d6+d44_out_r2n",d6.shape) #d6+d44_out_r2n (?, 32, 32, 32, 64)
-
-# plot_buf_1= tf.reshape(d6, [-1, 32, 32, 4]) ##############################
-# tf.summary.image("d6+d44_out_de", plot_buf_1)#############################
-
- d7 = tools.Ops.xxlu(tools.Ops.deconv3d(d6, k=3, out_c=de_c[4], str=1, name='de_c7'), label='lrelu')
- print("d7_out_r2n",d7.shape) #d7_out_r2n (?, 32, 32, 32, 32)
- d8 = tools.Ops.xxlu(tools.Ops.deconv3d(d7, k=3, out_c=de_c[4], str=1, name='de_c8'), label='lrelu')
- print("d8_out_r2n",d8.shape)#d8_out_r2n (?, 32, 32, 32, 32)
- d77 = tools.Ops.xxlu(tools.Ops.deconv3d(d7, k=3, out_c=de_c[4], str=1, name='de_c77'), label='lrelu')
- print("d77_out_r2n",d77.shape)#d77_out_r2n (?, 32, 32, 32, 32)
- d8 = d8 + d77
- print("d8+d77_out_r2n",d8.shape) #d8+d77_out_r2n (?, 32, 32, 32, 32)
-
-# plot_buf_1= tf.reshape(d8, [-1, 32, 32, 4]) ##############################
-# tf.summary.image("d8+d77_out_de", plot_buf_1)#############################
-
- d11 = tools.Ops.deconv3d(d8, k=3, out_c=de_c[5], str=1, name='de_c9')
- print("d11_out_r2n",d11.shape) #d11_out_r2n (?, 32, 32, 32, 1)
-
-# plot_buf_1= tf.reshape(d11, [-1, 32, 32,4]) ##########################
-# tf.summary.image("Ref_input", plot_buf_1)#############################
-
- ref_in = tf.reshape(d11, [-1, vox_res32, vox_res32, vox_res32],name="ref_in") ###
-
- y = tf.nn.sigmoid(d11,name='de_sigmoid')
-
- att_o = tf.reshape(y, [-1, vox_res32, vox_res32, vox_res32],name="de_out")
-
- print("att_out_shape",att_o.shape) #att_out_shape (?, 32, 32, 32)
-
- with tf.variable_scope('ref_net'):
-
- ref_o=refiner_network(ref_in)
-
-# plot_buf_1= tf.reshape(ref_o, [-1, 32, 32,4]) ######################
-# tf.summary.image("Ref_Out", plot_buf_1)#############################
-
- return ref_o,att_o, weights
-
- def build_graph(self):
- img_res = 127
- vox_res = 32
- self.X_rgb = tf.placeholder(shape=[None, None, img_res, img_res, 3], dtype=tf.float32)
- self.Y_vox = tf.placeholder(shape=[None, vox_res, vox_res, vox_res], dtype=tf.float32)
- self.lr = tf.placeholder(tf.float32)
- self.refine_lr = tf.placeholder(tf.float32)
- with tf.device('/cpu:0'):
- self.Y_pred,self.vae_o, self.weights = self.base_r2n2(self.X_rgb)
- tf.summary.histogram('Attsets_Weights', self.weights)
- with tf.device('/cpu:0'):
- ### rec loss
- print ("reached")
- with tf.variable_scope('Loss_Fun'):
- Y_vox_ = tf.reshape(self.Y_vox, shape=[-1, vox_res ** 3])
- Y_pred_ = tf.reshape(self.Y_pred, shape=[-1, vox_res ** 3])
- vae_o_=tf.reshape(self.vae_o, shape=[-1, vox_res ** 3])
-
- self.vae_loss = tf.reduce_mean(-tf.reduce_mean(Y_vox_ * tf.log(vae_o_ + 1e-8), reduction_indices=[1]) -
- tf.reduce_mean((1 - Y_vox_) * tf.log(1 - vae_o_ + 1e-8),reduction_indices=[1]))
- self.rec_loss = tf.reduce_mean(-tf.reduce_mean(Y_vox_ * tf.log(Y_pred_ + 1e-8), reduction_indices=[1]) -
- tf.reduce_mean((1 - Y_vox_) * tf.log(1 - Y_pred_ + 1e-8),reduction_indices=[1]))
- sum_rec_loss = tf.summary.scalar('rec_loss', self.rec_loss)
- self.sum_merged = sum_rec_loss
- tf.summary.histogram('rec_loss', self.rec_loss)
- tf.summary.scalar("vae-loss",self.vae_loss)
- tf.summary.histogram("vae_loss",self.vae_loss)
- self.mean_loss = tf.div(x=tf.math.add(x=self.vae_loss,y=self.rec_loss,name='add_loss'),y=2,name='mean_loss')
- tf.summary.histogram("mean_vae_loss",self.mean_loss)
- tf.summary.scalar("mean_vae_loss",self.mean_loss)
-
- with tf.variable_scope('Evaluation_Metric'):
- gt_vox=Y_vox_
- self.iou_ref = metric_iou(Y_pred_,gt_vox)
- tf.summary.scalar('iou_refiner', self.iou_ref)
- tf.summary.histogram('iou_refiner', self.iou_ref)
- self.iou_vae = metric_iou(vae_o_,gt_vox)
- tf.summary.scalar('iou_vae', self.iou_vae)
- tf.summary.histogram("iou_vae",self.iou_vae)
-
- with tf.variable_scope('Optimization'):
-
- base_en_var = [var for var in tf.trainable_variables() if var.name.startswith('Encoder/en')]
- base_dec_var = [var for var in tf.trainable_variables() if var.name.startswith('Decoder/de')]
- att_var = [var for var in tf.trainable_variables() if var.name.startswith('Att_Net/att')]
- refine_var = [var for var in tf.trainable_variables() if var.name.startswith('ref_net/ref')]
- self.refine_optim = tf.train.AdamOptimizer(learning_rate=self.refine_lr).minimize(self.rec_loss, var_list=refine_var)
- self.base_en_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=base_en_var)
- self.base_de_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=base_dec_var)
- self.att_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=att_var)
-
-
- print ("total weights:",tools.Ops.variable_count())
- self.saver = tf.train.Saver(max_to_keep=1)
-
- config = tf.ConfigProto(allow_soft_placement=True)
- #config.gpu_options.allow_growth = True
- #config.gpu_options.visible_device_list = '0,1'
-
- self.sess = tf.Session(config=npu_config_proto(config_proto=config))
- self.merged = tf.summary.merge_all()
- self.sum_writer_train = tf.summary.FileWriter(self.train_sum_dir, self.sess.graph)
- self.sum_writer_test = tf.summary.FileWriter(self.test_sum_dir, self.sess.graph)
-
- #######################
- path = self.train_mod_dir
- #path = './Model_released/' # retrain the released model
-
- if os.path.isfile(path + 'model.cptk.data-00000-of-00001'):
- print ("restoring saved model!")
- self.saver.restore(self.sess, path + 'model.cptk')
- else:
- self.sess.run(tf.global_variables_initializer())
-
-
- return 0
-
- def train(self, data):
- for epoch in range(0, 20, 1):
- train_view_num = 24 ##!!!!!!!!!!!
- data.shuffle_train_files(epoch, train_mv=train_view_num)
- total_train_batch_num = data.total_train_batch_num #int(len(self.X_rgb_train_files)/(self.batch_size*train_mv))
- print ('total_train_batch_num:', total_train_batch_num)
- for i in range(total_train_batch_num):
- #### training
- X_rgb_bat, Y_vox_bat = data.load_X_Y_train_next_batch(train_mv=train_view_num)
- print("multi_view_train_X_rgb_bat : ",X_rgb_bat.shape)#np.asarray(X.append(X_rgb[b*train_mv:(b+1)*train_mv,:]))
-
-
- print(time.ctime())
-
- ##### option 1: seperate train, seperate optimize
- #if epoch<=30:
- # single_view_train=True
- # multi_view_train=False
- #else:
- # single_view_train=False
- # multi_view_train=True
-
- ##### optiion 2: joint train, seperate optimize
- single_view_train = True
- multi_view_train = False
-
- if epoch <= 5:
- att_lr=.0002
- ref_lr=.0002
- if epoch > 5 and epoch <= 40:
- att_lr=.0001
- ref_lr=.0001
- if epoch > 40:
- att_lr=.00005
- ref_lr=.00005
- ########### single view train
- if single_view_train:
-
- rgb = np.reshape(X_rgb_bat,[batch_size*train_view_num, 1, 127,127,3])
- print("single_view_train_rgb_input_shape ",rgb.shape)
- vox = np.tile(Y_vox_bat[:,None,:,:,:],[1,train_view_num,1,1,1])
- vox = np.reshape(vox, [batch_size*train_view_num, 32,32,32])
- start_time=time.time()
- vae_loss_c,eee,ddd, rec_loss_c, sum_train,rrr,mean_vae,iou_ref_,iou_vae_ = self.sess.run([self.vae_loss,self.base_en_optim2,self.base_de_optim2,self.rec_loss,self.merged,self.refine_optim,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr,self.refine_lr: ref_lr})
- print('TimeHistory = %.4f' % (time.time()-start_time))
- print ('ep:', epoch, 'i:', i, 'train single rec loss:', rec_loss_c)
- print ('ep:', epoch, 'i:', i, 'train single vae loss:', vae_loss_c)
- print ('ep:', epoch, 'i:', i, 'train single mean_vae loss:',mean_vae)
- print ('ep:', epoch, 'i:', i, 'train single ref_iou:',iou_ref_)
- print ('ep:', epoch, 'i:', i, 'train single vae_iou:',iou_vae_)
-
- ########## multi view train
- if multi_view_train:
-
- vae_loss_c,rec_loss_c, _, sum_train,xxx,mean_vae,iou_ref_,iou_vae_ = self.sess.run([self.vae_loss,self.rec_loss, self.att_optim2, self.merged,self.refine_optim,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: X_rgb_bat, self.Y_vox: Y_vox_bat,self.lr: att_lr,self.refine_lr: ref_lr})
- print ('ep:', epoch, 'i:', i, 'train multi rec loss:', rec_loss_c)
- print ('ep:', epoch, 'i:', i, 'train multi vae loss:', vae_loss_c)
- print('ep:',epoch,'i',i,'train multi mean_vae loss:',mean_vae)
- print ('ep:', epoch, 'i:', i, 'train multi ref_iou:',iou_ref_)
- print ('ep:', epoch, 'i:', i, 'train multi vae_iou:',iou_vae_)
-
- ############
- if i % 5 == 0:
- self.sum_writer_train.add_summary(sum_train, epoch * total_train_batch_num + i)
-
-
- #### testing
- if i % 10 == 0 :
- X_rgb_batch, Y_vox_batch = data.load_X_Y_test_next_batch(test_mv=3)
-
-# vae_pred = tf.get_default_graph().get_tensor_by_name("Decoder/de_out:0")
-# ref_pred = tf.get_default_graph().get_tensor_by_name("ref_net/ref_Dec/ref_out:0")
-# gt_vox=Y_vox_batch.astype(np.float32)
-
-# iou_pred = metric_iou(ref_pred,gt_vox)
-# tf.summary.scalar("iou",iou_pred)
-
- rrrr,aaaa,rec_loss_te, qwerty, Y_vox_test_pred, att_pred, sum_test,mean_vae,iou_ref_,iou_vae_ = \
- self.sess.run([self.refine_optim,self.att_optim2,self.rec_loss,self.vae_loss, self.Y_pred,self.weights, self.merged,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: X_rgb_batch, self.Y_vox: Y_vox_batch,self.lr: 0,self.refine_lr: 0})
-
- X_rgb_batch = X_rgb_batch.astype(np.float16)
-
- Y_vox_batch = Y_vox_batch.astype(np.float16)
- Y_vox_test_pred = Y_vox_test_pred.astype(np.float16)
- att_pred = att_pred.astype(np.float16)
- to_save = {'X_test':X_rgb_batch,'Y_test_pred':Y_vox_test_pred,'att_pred':att_pred,'Y_test_true':Y_vox_batch}
- scipy.io.savemat(self.test_res_dir+'X_Y_pred_'+str(epoch).zfill(2)+'_'+str(i).zfill(5)+'.mat',to_save,do_compression=True)
-
-
- self.sum_writer_test.add_summary(sum_test, epoch * total_train_batch_num + i)
-
-# iou_ref=evaluate_voxel_prediction(ref_pred_,gt_vox)
-# iou_vae=evaluate_voxel_prediction(vae_pred_,gt_vox)
-
-# print("Ref_iou:",iou_ref)
-# print("Vae_iou:",iou_vae)
-
-# plot_list_iou.append(iou_ref)
-# plot_list_i.append((i/50))
-# graph_plot(plot_list_iou,plot_list_i)
- print ('ep:', epoch, 'i:', i, 'test rec loss:', rec_loss_te)
- print ('ep:', epoch, 'i:', i, 'test vae loss:', qwerty )
- print ('ep:', epoch, 'i:', i, 'test mean_vae loss:', mean_vae)
- print ('ep:', epoch, 'i:', i, 'test ref_iou:',iou_ref_)
- print ('ep:', epoch, 'i:', i, 'test vae_iou:',iou_vae_)
-
- #### model saving
- #if i % 100 == 0 :
- #self.saver.save( self.sess, save_path=self.train_mod_dir + 'model.cptk' )
- #print ( 'epoch:', epoch, 'i:', i, 'model saved!' )
-
-
-# plt.show()
-
-
-##########
-if __name__ =='__main__':
- npu_keras_sess = set_keras_session_npu_config()
-
- net = Network() #net=object to create instance
-
- print("network compleated") ###
-
- net.build_graph()
- print("graph compleated")
-
-
- data = tools.Data(config)
- print("tools.data compleated")
-
-
- print('training data')
-
- net.train(data)
-
- close_session(npu_keras_sess)
-
-
-
-
-
-
-
-
-
-
-
-
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/modelarts_entry.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/modelarts_entry.py
deleted file mode 100644
index 4fb8aca77..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/modelarts_entry.py
+++ /dev/null
@@ -1,55 +0,0 @@
-#
-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-
-import os
-import argparse
-import sys
-
-# 解析输入参数data_url
-parser = argparse.ArgumentParser()
-parser.add_argument("--data_url", type=str, default="/cache/dataset")
-parser.add_argument("--train_url", type=str, default="/cache/output")
-config = parser.parse_args()
-
-print("[CANN-Modelzoo] code_dir path is [%s]" % (sys.path[0]))
-code_dir = sys.path[0]
-
-print("[CANN-Modelzoo] work_dir path is [%s]" % (os.getcwd()))
-work_dir = os.getcwd()
-
-print("[CANN-Modelzoo] start run train shell")
-# 执行训练脚本
-shell_cmd = ("bash %s/npu_train.sh %s %s %s %s " % (code_dir, code_dir, work_dir, config.data_url, config.train_url))
-os.system(shell_cmd)
-print("[CANN-Modelzoo] finish run train shell")
-
-
-
-
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/npu_train.sh b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/npu_train.sh
deleted file mode 100644
index a91aac698..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/npu_train.sh
+++ /dev/null
@@ -1,59 +0,0 @@
-#!/bin/bash
-
-code_dir=$1
-work_dir=$2
-dataset_path=$3
-output_path=$4
-
-#############训练前输入目录文件确认#########################
-echo "[CANN-Modelzoo] before train - list my run files[/usr/local/Ascend/ascend-toolkit]:"
-ls -al /usr/local/Ascend/ascend-toolkit
-echo ""
-
-echo "[CANN-Modelzoo] before train - list my code files[${code_dir}]:"
-ls -al ${code_dir}
-echo ""
-
-echo "[CANN-Modelzoo] before train - list my work files[${work_dir}]:"
-ls -al ${work_dir}
-echo ""
-
-echo "[CANN-Modelzoo] before train - list my dataset files[${dataset_path}]:"
-ls -al ${dataset_path}
-echo ""
-
-echo "[CANN-Modelzoo] before train - list my output files[${output_path}]:"
-ls -al ${output_path}
-echo ""
-
-######环境变量修改######
-###如果需要修改环境变量的，在此处修改
-#设置日志级别为info
-#export ASCEND_GLOBAL_LOG_LEVEL=1
-#设置日志打屏到屏幕
-#export ASCEND_SLOG_PRINT_TO_STDOUT=1
-#export TF_CPP_MIN_LOG_LEVEL=0
-env > ${output_path}/my_env.log
-
-######训练执行######
-###此处每个网络执行命令不同，需要修改
-python3.7 ${code_dir}/main_AttSets.py --data_url=${dataset_path} --train_url=${output_path}
-if [ $? -eq 0 ];
-then
- echo "[CANN-Modelzoo] train return success"
-else
- echo "[CANN-Modelzoo] train return failed"
-fi
-
-######训练后把需要备份的内容保存到output_path######
-###此处每个网络不同，视情况添加cp
-cp -r ${work_dir} ${output_path}
-
-######训练后输出目录文件确认######
-echo "[CANN-Modelzoo] after train - list my work files[${work_dir}]:"
-ls -al ${work_dir}
-echo ""
-
-echo "[CANN-Modelzoo] after train - list my output files[${output_path}]:"
-ls -al ${output_path}
-echo ""
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/ops_info.json b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/ops_info.json
deleted file mode 100644
index 6d5d0ba20..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/ops_info.json
+++ /dev/null
@@ -1,10 +0,0 @@
-{
- "black-list": {
- "to-add": [
- "Mul",
- "ReduceSumD",
- "SquaredDifference"
- ]
- }
-}
-
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/tools.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/tools.py
deleted file mode 100644
index 6054ceb1e..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/tools.py
+++ /dev/null
@@ -1,482 +0,0 @@
-#
-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-from npu_bridge.npu_init import *
-import tensorflow as tf
-from PIL import Image
-import numpy as np
-import os
-import sys
-import random
-import matplotlib.pyplot as plt
-import matplotlib.gridspec as gridspec
-from mpl_toolkits import mplot3d
-import re
-import binvox_rw as binvox_rw
-#################################################################
-from voxel import voxel2obj
-#from PyMCubes import mcubes
-from export_obj import export_obj
-#################################################################
-class Data:
- def __init__(self, config):
- self.config = config
- self.batch_size = config['batch_size']
- self.train_batch_index = 0
- self.test_batch_index_sq = 0
- self.cat_names = config['cat_names']
- self.total_mv = config['total_mv']
- self.cat_test_1st_index = None
-
- self.X_rgb_train_files_ori, self.Y_vox_train_files_ori, self.X_rgb_test_files_ori, self.Y_vox_test_files_ori \
- =self.load_X_Y_files_paths_all(self.cat_names)
-
- print ('X_rgb_train_files_ori:', len(self.X_rgb_train_files_ori))
- print ('X_rgb_test_files_ori:', len(self.X_rgb_test_files_ori))
-
- @staticmethod
- def crop_center(im, new_height, new_width):
-# print('crop_center')
- height = im.shape[0] # Get dimensions
- width = im.shape[1]
- left = (width - new_width) // 2
- top = (height - new_height) // 2
- right = (width + new_width) // 2
- bottom = (height + new_height) // 2
- return im[top:bottom, left:right]
-
- @staticmethod
- def add_random_color_background(im, color_range):
-# print('add_random_color_background')
- r, g, b = [np.random.randint(color_range[i][0], color_range[i][1] + 1) for i in range(3)]
-# print (r)
-# print (g)
-# print (b)
- if isinstance(im, Image.Image):
- im = np.array(im)
-# print (im.shape)
-# b = np.array([3]) ###
-# np.concatenate((im, '3')) ##
-# im=np.append(im,3) ###
-# print (im.shape) ###
- if im.shape[2] > 3:
-
- # If the image has the alpha channel, add the background
- alpha = (np.expand_dims(im[:, :, 3], axis=2) == 0).astype(np.float)
-
- im = im[:, :, :3]
- bg_color = np.array([[[r, g, b]]])
- im = alpha * bg_color + (1 - alpha) * im
- return im
-
- @staticmethod
- def image_transform(img, crop_x, crop_y, crop_loc=None, color_tint=None):
-# print('image_transform')
- RANDOM_CROP = True
- # Slight translation
- if RANDOM_CROP and not crop_loc:
- crop_loc = [np.random.randint(0, crop_y), np.random.randint(0, crop_x)]
-
- if crop_loc:
- cr, cc = crop_loc
- height, width, _ = img.shape
- img_h = height - crop_y
- img_w = width - crop_x
- img = img[cr:cr + img_h, cc:cc + img_w]
- # depth = depth[cr:cr+img_h, cc:cc+img_w]
-
- FLIP = True
- if FLIP and np.random.rand() > 0.5:
- img = img[:, ::-1, ...]
- return img
-
- @staticmethod
- def preprocess_img(im, train):
-# print('processing_img')
- # add random background
- TRAIN_NO_BG_COLOR_RANGE = [[225, 255], [225, 255], [225, 255]]
- TEST_NO_BG_COLOR_RANGE = [[240, 240], [240, 240], [240, 240]]
- im = Data.add_random_color_background(im, TRAIN_NO_BG_COLOR_RANGE if train else TEST_NO_BG_COLOR_RANGE)
-
- # If the image has alpha channel, remove it.
- CONST_IMG_W = 127
- CONST_IMG_H = 127
- im_rgb = np.array(im)[:, :, :3].astype(np.float32)
- if train:
- # Data augmentation
- PAD_X = 10
- PAD_Y = 10
- t_im = Data.image_transform(im_rgb, PAD_X, PAD_Y)
- else:
- t_im = Data.crop_center(im_rgb, CONST_IMG_H, CONST_IMG_W)
-
- # Scale image
- t_im = t_im / 255.
- return t_im
-
- @staticmethod
- def load_single_X_rgb_r2n2(img_path, train):
-# print('load_single_X_rgb_r2n2')
- im = Image.open(img_path)
-
- t_im = Data.preprocess_img(im, train=train)
- # plt.figure()
- # plt.imshow(t_im)
- return t_im
-
- @staticmethod
- def load_single_Y_vox(vox_path):
-# print('load_single_Y_vox')
- with open(vox_path, 'rb') as ff:
- vox = binvox_rw.read_as_3d_array(ff)
- vox_grid = vox.data.astype(int)
-
- #Data.plotFromVoxels(vox_grid)
- return vox_grid
-
- @staticmethod
- def load_X_Y_files_paths(X_cat_folder, Y_cat_folder):
-# print('load_X_Y_files_paths')
- X_obj_folders=[X_f for X_f in sorted(os.listdir(X_cat_folder))]
- Y_obj_folders=[Y_f for Y_f in sorted(os.listdir(Y_cat_folder))]
- if len(X_obj_folders) != len(Y_obj_folders):
- print ('Files are inconsistent in:', X_cat_folder, 'and', Y_cat_folder)
-
- #### split train/test, according to 3D-R2N2 paper
- train_num = int(0.8*len(X_obj_folders))
- idx = list(range(train_num))
-
- test_objs = []
- X_train_obj_folders =[]; Y_train_obj_folders=[]
- X_test_obj_folders =[]; Y_test_obj_folders=[]
- for i in range(len(X_obj_folders)):
- obj_na = X_obj_folders[i]
- if obj_na not in Y_obj_folders:
- print ('inconsistent single obj ignored')
- continue
- if i in idx:
- X_train_obj_folders.append(X_cat_folder+obj_na+'/')
- Y_train_obj_folders.append(Y_cat_folder+obj_na+'/')
- else:
- X_test_obj_folders.append(X_cat_folder+obj_na+'/')
- Y_test_obj_folders.append(Y_cat_folder+obj_na+'/')
- test_objs.append(obj_na)
- print ('train objs:', len(X_train_obj_folders))
- print ('test objs:', len(X_test_obj_folders))
-
- #########
- def load_x_y_files(X_obj_fo, Y_obj_fo):
-# print('load_x_y_files')
- X_files_paths = []
- Y_files_paths = []
- for j in range(len(X_obj_fo)):
- if X_obj_fo[j][-5:] not in Y_obj_fo[j]:
- print ('inconsistent single obj exit')
- exit()
- for xf in sorted(os.listdir(X_obj_fo[j]+'rendering/')):
- if '.png' in xf:
- X_files_paths.append(X_obj_fo[j]+'rendering/'+xf)
- Y_files_paths.append(Y_obj_fo[j]+'model.binvox')
- return X_files_paths, Y_files_paths
- #########
-
- X_train_files, Y_train_files = load_x_y_files(X_train_obj_folders, Y_train_obj_folders)
- X_test_files, Y_test_files = load_x_y_files(X_test_obj_folders, Y_test_obj_folders)
-
- return X_train_files, Y_train_files,X_test_files,Y_test_files
-
- @staticmethod
- def load_X_Y_rgb_vox(X_files_full_path, Y_files_full_path, train):
-# print('load_X_Y_rgb_vox')
- if len(X_files_full_path) != len(Y_files_full_path):
- print ('load_X_Y_rgb_vox error!')
- exit()
- X_rgb = []
- Y_vox = []
- for X_f, Y_f in zip(X_files_full_path, Y_files_full_path):
- na = re.split('/', X_f)[-3]
- if na not in Y_f:
- print ('X Y rgb vox file not consistent!')
- exit()
- rgb = Data.load_single_X_rgb_r2n2(X_f, train=train)
- X_rgb.append(rgb)
-
- vox = Data.load_single_Y_vox(Y_f)
- Y_vox.append(vox)
-
- X_rgb = np.asarray(X_rgb)
- Y_vox = np.asarray(Y_vox)
- return X_rgb, Y_vox
-
- def test_listdir(self, path):
- print('当前执行路径', os.getcwd())
- for file in os.listdir(path):
- file_path = os.path.join(path, file)
- print(file_path)
-
- def load_X_Y_files_paths_all(self,cat_names):
-# print('load_X_Y_files_paths_all')
- x_rgb_str='X_rgb_'
- y_vox_str='Y_vox_'
-
- X_train_files_paths_all=[]
- Y_trian_files_paths_all=[]
- X_test_files_paths_all=[]
- Y_test_files_paths_all=[]
- self.cat_test_1st_index = [0]
- for name in cat_names:
- self.test_listdir(os.getcwd())
- print ('loading files:', name)
- X_rgb_folder = self.config[x_rgb_str+name]
- Y_vox_folder = self.config[y_vox_str+name]
-
- X_train_files, Y_train_files, X_test_files, Y_test_files = self.load_X_Y_files_paths(X_rgb_folder, Y_vox_folder)
- self.cat_test_1st_index.append(len(X_test_files))
-
- for X_rgb_f, Y_vox_f in zip(X_train_files, Y_train_files):
- X_train_files_paths_all.append(X_rgb_f)
- Y_trian_files_paths_all.append(Y_vox_f)
- for X_rgb_f, Y_vox_f in zip(X_test_files, Y_test_files):
- X_test_files_paths_all.append(X_rgb_f)
- Y_test_files_paths_all.append(Y_vox_f)
-
- return X_train_files_paths_all, Y_trian_files_paths_all,X_test_files_paths_all,Y_test_files_paths_all
-
- ################################
- def load_X_Y_train_next_batch(self, train_mv):
-# print('load_X_Y_train_next_batch')
- X_rgb_files = self.X_rgb_train_files[self.batch_size*self.train_batch_index*train_mv:self.batch_size*(self.train_batch_index+1)*train_mv]
- Y_vox_files = self.Y_vox_train_files[self.batch_size*self.train_batch_index*train_mv:self.batch_size*(self.train_batch_index+1)*train_mv]
- self.train_batch_index +=1
-
- X_rgb, Y_vox = self.load_X_Y_rgb_vox(X_rgb_files, Y_vox_files, train=True)
- X = []
- Y = []
- for b in range(self.batch_size):
- X.append(X_rgb[b * train_mv:(b + 1) * train_mv, :])
- Y.append(Y_vox[b * train_mv, :, :, :])
- X = np.asarray(X)
- Y = np.asarray(Y)
- return X, Y
-
- def load_X_Y_test_next_batch(self, test_mv):
-# print('load_X_Y_test_next_batch')
- num = self.total_mv
- idx = random.sample(range(len(self.X_rgb_test_files_ori)//num), self.batch_size) ##############added extra '/'
- X_rgb_files = []
- Y_vox_files =[]
-
- for i in idx:
- tp1 = self.X_rgb_test_files_ori[i*num:(i+1)*num]
- tp2 = self.Y_vox_test_files_ori[i*num:(i+1)*num]
- for x, y in zip(tp1, tp2):
- X_rgb_files.append(x)
- Y_vox_files.append(y)
- X_rgb_batch, Y_vox_batch = self.load_X_Y_rgb_vox(X_rgb_files, Y_vox_files, train=False)
-
- X =[]
- Y =[]
- for b in range(self.batch_size):
- tp1 = X_rgb_batch[b*num:(b+1)*num,:,:,:]
- tp2 = Y_vox_batch[b*num:(b+1)*num,:,:,:]
- idx2 = random.sample(range(num), test_mv)
- X.append(tp1[idx2])
- Y.append(tp2[idx2[0]])
- X = np.asarray(X)
- Y = np.asarray(Y)
- return X, Y
-
- def shuffle_train_files(self, ep, train_mv):
-# print('shuffle_train_files')
- num = self.total_mv
- X_rgb_new=[]; Y_vox_new=[]
- self.train_batch_index = 0
- X_rgb = self.X_rgb_train_files_ori
- Y_vox = self.Y_vox_train_files_ori
- index = list(range(int(len(X_rgb)/num)))
- random.Random(ep).shuffle(index)
-
- for i in index:
- tp1 = X_rgb[i*num:(i+1)*num]
- tp2 = Y_vox[i*num:(i+1)*num]
-
- view_ind = list(range(num))
- random.Random(i+100).shuffle(view_ind)
- valid_view_num = int(num / train_mv) * train_mv
- view_ind = view_ind[0:valid_view_num]
-
- for j in view_ind:
- x = tp1[j]
- y = tp2[j]
- na = re.split('/', x)[-3]
- if na not in y:
- print ('X Y rgb vox file not consistent!')
- exit()
- X_rgb_new.append(x)
- Y_vox_new.append(y)
- self.X_rgb_train_files = X_rgb_new
- self.Y_vox_train_files = Y_vox_new
- self.total_train_batch_num = int(len(self.X_rgb_train_files)/(self.batch_size*train_mv))
-
- def shuffle_test_files(self, test_mv, seed):
-# print('shuffle_test_files')
- X_rgb_new=[]; Y_vox_new=[]
- X_rgb = self.X_rgb_test_files_ori
- Y_vox = self.Y_vox_test_files_ori
-
- num = self.total_mv
- ###
- cat_test_1st_index_new =[]
- for k in self.cat_test_1st_index:
- cat_test_1st_index_new.append( int(k/num)* int(num/test_mv)*test_mv)
- self.cat_test_1st_index=cat_test_1st_index_new
- ###
- total_obj = int(len(X_rgb)/num)
- for i in range(total_obj):
- view_ind = list(range(num))
- random.Random(i+seed).shuffle(view_ind)
-
- valid_view_num = int(num/test_mv)*test_mv
- view_ind = view_ind[0:valid_view_num]
- for id in view_ind:
- X_rgb_new.append(X_rgb[i*num+id])
- Y_vox_new.append(Y_vox[i*num+id])
-
- self.X_rgb_test_files = X_rgb_new
- self.Y_vox_test_files = Y_vox_new
-
-class Ops:
-
- @staticmethod
- def lrelu(x, leak=0.2):
- return tf.nn.leaky_relu(x,alpha=0.2,name=None)
-
- @staticmethod
- def relu(x):
- return tf.nn.relu(x)
-
- @staticmethod
- def xxlu(x, label='relu'):
- if label == 'relu':
- return Ops.relu(x)
- if label == 'lrelu':
- return Ops.lrelu(x, leak=0.2)
-
- @staticmethod
- def variable_sum(var, name):
- with tf.name_scope(name):
- mean = tf.reduce_mean(var)
- tf.summary.scalar('mean', mean)
- stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
- tf.summary.scalar('stddev', stddev)
- tf.summary.scalar('max', tf.reduce_max(var))
- tf.summary.scalar('min', tf.reduce_min(var))
- tf.summary.histogram('histogram', var)
-
- @staticmethod
- def variable_count():
- total_para = 0
- for variable in tf.trainable_variables():
- shape = variable.get_shape()
- variable_para = 1
- for dim in shape:
- variable_para *= dim.value
- total_para += variable_para
- return total_para
-
- @staticmethod
- def fc(x, out_d, name):
- xavier_init = tf.contrib.layers.xavier_initializer()
- zero_init = tf.zeros_initializer()
- in_d = x.get_shape()[1]
- w = tf.get_variable(name + '_w', [in_d, out_d], initializer=xavier_init,dtype=tf.float32)
- b = tf.get_variable(name + '_b', [out_d], initializer=zero_init,dtype=tf.float32)
- y = tf.nn.bias_add(tf.matmul(x, w), b)
- Ops.variable_sum(w, name)
- return y
-
- @staticmethod
- def conv2d(x, k, out_c, str, name, pad='SAME'):
- xavier_init = tf.contrib.layers.xavier_initializer()
- zero_init = tf.zeros_initializer()
- in_c = x.get_shape()[3]
- w = tf.get_variable(name + '_w', [k, k, in_c, out_c], initializer=xavier_init, dtype=tf.float32)
- b = tf.get_variable(name + '_b', [out_c], initializer=zero_init,dtype=tf.float32)
-
- stride = [1, str, str, 1]
- y = tf.nn.bias_add(tf.nn.conv2d(x, w, stride, pad), b)
- Ops.variable_sum(w, name)
- return y
-
- @staticmethod
- def maxpool2d(x, k, s,name, pad='SAME'):
- ker = [1,k,k,1]
- str = [1,s,s,1]
- y = tf.nn.max_pool(x,ksize=ker, strides=str, padding=pad, name=name)
- return y
-
- @staticmethod
- def conv3d(x, k, out_c, str, name, pad='SAME'):
- xavier_init = tf.contrib.layers.xavier_initializer()
- zero_init = tf.zeros_initializer()
- in_c = x.get_shape()[4]
- w = tf.get_variable(name + '_w', [k, k, k, in_c, out_c], initializer=xavier_init, dtype=tf.float32)
- b = tf.get_variable(name + '_b', [out_c], initializer=zero_init, dtype=tf.float32)
- stride = [1, str, str, str, 1]
- y = tf.nn.bias_add(tf.nn.conv3d(x, w, stride, pad), b)
- Ops.variable_sum(w, name)
- return y
-
- @staticmethod
- def deconv3d(x, k, out_c, str, name, pad='SAME'):
- xavier_init = tf.contrib.layers.xavier_initializer()
- zero_init = tf.zeros_initializer()
- bat = tf.shape(x)[0]
- [_, in_d1, in_d2, in_d3, in_c] = x.get_shape()
- in_d1 = int(in_d1); in_d2 = int(in_d2); in_d3 = int(in_d3); in_c = int(in_c)
- w = tf.get_variable(name + '_w', [k, k, k, out_c, in_c], initializer=xavier_init, dtype=tf.float32)
- b = tf.get_variable(name + '_b', [out_c], initializer=zero_init, dtype=tf.float32)
-
- out_shape = [bat, in_d1 * str, in_d2 * str, in_d3 * str, out_c]
- stride = [1, str, str, str, 1]
- y = tf.nn.conv3d_transpose(x, w, output_shape=out_shape, strides=stride, padding=pad)
- y = tf.nn.bias_add(y, b)
- Ops.variable_sum(w, name)
- return y
-
- @staticmethod
- def maxpool3d(x,k,s,name,pad='SAME'):
- ker = [1,k,k,k,1]
- str = [1,s,s,s,1]
- y = tf.nn.max_pool3d(x,ksize=ker,strides=str,padding=pad, name=name)
- return y
-
-
-
-
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/voxel.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/voxel.py
deleted file mode 100644
index 8bce75ecf..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Code/voxel.py
+++ /dev/null
@@ -1,89 +0,0 @@
-#
-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-from npu_bridge.npu_init import *
-import numpy as np
-
-
-def evaluate_voxel_prediction(preds, gt, thresh):
- preds_occupy = preds[:, 1, :, :] >= thresh
- diff = np.sum(np.logical_xor(preds_occupy, gt[:, 1, :, :]))
- intersection = np.sum(np.logical_and(preds_occupy, gt[:, 1, :, :]))
- union = np.sum(np.logical_or(preds_occupy, gt[:, 1, :, :]))
- num_fp = np.sum(np.logical_and(preds_occupy, gt[:, 0, :, :])) # false positive
- num_fn = np.sum(np.logical_and(np.logical_not(preds_occupy), gt[:, 1, :, :])) # false negative
- return np.array([diff, intersection, union, num_fp, num_fn])
-
-
-def voxel2mesh(voxels, surface_view):
- cube_verts = [[0, 0, 0], [0, 0, 1], [0, 1, 0], [0, 1, 1], [1, 0, 0], [1, 0, 1], [1, 1, 0],
- [1, 1, 1]] # 8 points
-
- cube_faces = [[0, 1, 2], [1, 3, 2], [2, 3, 6], [3, 7, 6], [0, 2, 6], [0, 6, 4], [0, 5, 1],
- [0, 4, 5], [6, 7, 5], [6, 5, 4], [1, 7, 3], [1, 5, 7]] # 12 face
-
- cube_verts = np.array(cube_verts)
- cube_faces = np.array(cube_faces) + 1
-
- scale = 0.01
- cube_dist_scale = 1.1
- verts = []
- faces = []
- curr_vert = 0
-
- positions = np.where(voxels > 0.3)
- voxels[positions] = 1
- for i, j, k in zip(*positions):
- # identifies if current voxel has an exposed face
- if not surface_view or np.sum(voxels[i-1:i+2, j-1:j+2, k-1:k+2]) < 27:
- verts.extend(scale * (cube_verts + cube_dist_scale * np.array([[i, j, k]])))
- faces.extend(cube_faces + curr_vert)
- curr_vert += len(cube_verts)
-
- return np.array(verts), np.array(faces)
-
-
-def write_obj(filename, verts, faces):
- """ write the verts and faces on file."""
- with open(filename, 'w') as f:
- # write vertices
- f.write('g\n# %d vertex\n' % len(verts))
- for vert in verts:
- f.write('v %f %f %f\n' % tuple(vert))
-
- # write faces
- f.write('# %d faces\n' % len(faces))
- for face in faces:
- f.write('f %d %d %d\n' % tuple(face))
-
-
-def voxel2obj(filename, pred, surface_view = True):
- verts, faces = voxel2mesh(pred, surface_view)
- write_obj(filename, verts, faces)
-
--
Gitee
From 9aa28751ec8079a6f567fc616bb8b09f6f028936 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:41:27 +0000
Subject: [PATCH 02/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/Model?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/.keep | 0
.../contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/test_sum/.keep | 0
.../cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_mod/.keep | 0
.../cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_sum/.keep | 0
4 files changed, 0 insertions(+), 0 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/.keep
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/test_sum/.keep
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_mod/.keep
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_sum/.keep
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/.keep
deleted file mode 100644
index e69de29bb..000000000
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/test_sum/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/test_sum/.keep
deleted file mode 100644
index e69de29bb..000000000
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_mod/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_mod/.keep
deleted file mode 100644
index e69de29bb..000000000
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_sum/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/Model/train_sum/.keep
deleted file mode 100644
index e69de29bb..000000000
--
Gitee
From 3c739cc741fcd585484c940cb59bb34a449bb73c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:41:33 +0000
Subject: [PATCH 03/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/log?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/log/.keep | 0
1 file changed, 0 insertions(+), 0 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/log/.keep
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/log/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/log/.keep
deleted file mode 100644
index e69de29bb..000000000
--
Gitee
From f323ee0f5282945feaa4bb771dddf1ad1696c39c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:41:43 +0000
Subject: [PATCH 04/16] =?UTF-8?q?=E9=87=8D=E5=91=BD=E5=90=8D=20TensorFlow/?=
=?UTF-8?q?contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/test/=20train=5F?=
=?UTF-8?q?full=5F1p.sh=20=E4=B8=BA=20TensorFlow/contrib/cv/Pix2Vox=5FID12?=
=?UTF-8?q?84=5Ffor=5FTensorFlow/test/train=5Ffull=5F1p.sh?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../test/{ train_full_1p.sh => train_full_1p.sh} | 0
1 file changed, 0 insertions(+), 0 deletions(-)
rename TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/{ train_full_1p.sh => train_full_1p.sh} (100%)
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/ train_full_1p.sh b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
similarity index 100%
rename from TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/ train_full_1p.sh
rename to TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
--
Gitee
From 2447e38f0e503b3530c4a15b24efad2bc947aa34 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:41:49 +0000
Subject: [PATCH 05/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/test/.k?=
=?UTF-8?q?eep?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/.keep | 0
1 file changed, 0 insertions(+), 0 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/.keep
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/.keep
deleted file mode 100644
index e69de29bb..000000000
--
Gitee
From 0aca247b6f3d2817c6b1ec7446d1b8e2545db621 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:42:05 +0000
Subject: [PATCH 06/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/.keep?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/.keep | 0
1 file changed, 0 insertions(+), 0 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/.keep
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/.keep b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/.keep
deleted file mode 100644
index e69de29bb..000000000
--
Gitee
From b9d7198aed6bff338257a509fb03b4c2efcd2e64 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:42:17 +0000
Subject: [PATCH 07/16] update
TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_performance_1p.sh.
---
.../Pix2Vox_ID1284_for_TensorFlow/test/train_performance_1p.sh | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_performance_1p.sh b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_performance_1p.sh
index b1b6c36e1..ea30cc72d 100644
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_performance_1p.sh
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_performance_1p.sh
@@ -1,7 +1,7 @@
#!/bin/bash
#当前路径,不需要修改
-cur_path=`pwd`/../Code/
+cur_path=`pwd`/../
#集合通信参数,不需要修改
--
Gitee
From 281889d7370d53ed575b5f98be72cb9209df98cf Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:46:04 +0000
Subject: [PATCH 08/16] =?UTF-8?q?=E6=8F=90=E4=BA=A4=E8=AE=AD=E7=BB=83?=
=?UTF-8?q?=E7=9B=B8=E5=85=B3=E6=96=87=E4=BB=B6=20=E5=A2=9E=E5=8A=A0?=
=?UTF-8?q?=E6=96=87=E4=BB=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../main_AttSets.py | 765 ++++++++++++++++++
1 file changed, 765 insertions(+)
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/main_AttSets.py
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/main_AttSets.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/main_AttSets.py
new file mode 100644
index 000000000..69155def8
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/main_AttSets.py
@@ -0,0 +1,765 @@
+#
+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+from npu_bridge.npu_init import *
+import tensorflow as tf
+import os
+
+# os.system('pip install scipy==0.19.0')
+# os.system('pip install matplotlib==2.0.2')
+# os.system('pip install skimage==0.13.0')
+# os.system('pip install PyMCubes')
+
+import shutil
+import sys
+import scipy.io
+
+sys.path.append('..')
+import tools as tools
+import numpy as np
+import time
+from keras.layers import BatchNormalization, Conv3D, MaxPooling3D, Dense, Reshape, Add, LeakyReLU, Conv3DTranspose
+from keras.activations import relu, sigmoid, tanh
+from keras import models
+import copy
+import matplotlib.pyplot as plt
+
+import argparse
+
+# import moxing as mox
+
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--train_url', type=str, help='the path model saved')
+parser.add_argument("--data_url", type=str, default="./dataset")
+obs_config = parser.parse_args()
+
+###################
+# import tflearn
+###################
+
+batch_size = 2
+img_res = 127
+vox_res32 = 32
+total_mv = 24
+GPU0 = '0'
+GPU1 = '1'
+
+re_train = True
+# re_train=True
+single_view_train = True
+multi_view_train = True
+
+###########################
+
+plot_list_iou = []
+plot_list_i = []
+iii = 0
+config = {} # python dictionary
+config['batch_size'] = batch_size
+config['total_mv'] = total_mv
+# config['cat_names'] = ['02691156','02828884','04530566','03636649','03001627']
+# config['cat_names'] = ['02691156','02828884','02933112','02958343','03001627','03211117',
+# '03636649','03691459','04090263','04256520','04379243','04401088','04530566']
+config['cat_names'] = ['02691156']
+for name in config['cat_names']:
+ config['X_rgb_' + name] = obs_config.data_url + '/ShapeNetRendering/' + name + '/'
+
+ config['Y_vox_' + name] = obs_config.data_url + '/ShapeNetVox32/' + name + '/'
+
+
+# config['Y_vox_'+name] = '/home/wiproec4/3d reconstruction/attsets/Data_sample/shapenet dataset/ShapeNetVox32/train_1_dataset/'+name+'/'
+
+
+# output : {'batch_size': 1, 'total_mv': 24, 'cat_names': ['03001627'], 'Y_vox_03001627': '/home/wiproec4/3d reconstruction/attsets/Data_sample/#ShapeNetVox32/03001627/', 'X_rgb_03001627': '/home/wiproec4/3d reconstruction/attsets/Data_sample/ShapeNetRendering/03001627/'}
+
+def metric_iou(prediction, gt):
+ # labels = tf.greater_equal(gt[gt], 0.5)
+ # prediction = tf.cast(prediction,tf.int32)
+ predictions = tf.greater_equal(prediction, 0.5)
+ gt_ = tf.greater_equal(gt, 0.5)
+ intersection = tf.reduce_sum(tf.cast(tf.logical_and(predictions, gt_), tf.float32))
+ union = tf.reduce_sum(tf.cast(tf.math.logical_or(predictions, gt_), tf.float32))
+ iou = tf.cast(x=intersection, dtype=tf.float32) / tf.cast(x=union, dtype=tf.float32)
+ return iou
+
+
+def graph_plot(iou_value, i_value):
+ x = i_value
+ y = iou_value
+ plt.plot(x, y, color='green', linestyle='dashed', linewidth=3, marker='o', markerfacecolor='blue', markersize=12)
+
+ plt.ylim(0, 2)
+ plt.xlim(0, 500)
+
+ plt.xlabel('Iterations')
+
+ plt.ylabel('IOU')
+
+ plt.title('Refiner Accuracy')
+
+ plt.show()
+
+
+def evaluate_voxel_prediction(prediction, gt):
+ # """ The prediction and gt are 3 dim voxels. Each voxel has values 1 or 0"""
+ prediction[prediction >= 0.5] = 1
+ prediction[prediction < 0.5] = 0
+ intersection = np.sum(np.logical_and(prediction, gt))
+ union = np.sum(np.logical_or(prediction, gt))
+ IoU = float(intersection) / float(union)
+ return IoU
+
+
+#####################################
+def refiner_network(volumes_in):
+ with tf.device('/cpu:0'):
+ with tf.variable_scope('ref_enc'):
+ input_volumes_32 = tf.reshape(volumes_in, [-1, vox_res32, vox_res32, vox_res32, 1], name="ref_net_in")
+
+ print("input_volumes_32_shape", input_volumes_32.shape) # input_volumes_32_shape (?,32,32,32,1)
+
+ rn1 = Conv3D(filters=32, kernel_size=(4, 4, 4), padding='same', data_format="channels_last", name='ref_c1')(
+ input_volumes_32)
+ rn2 = BatchNormalization()(rn1)
+ rn3 = LeakyReLU(alpha=.2)(rn2)
+ print("rn3.shape", rn3.shape) # rn3.shape (?, 32, 32, 32, 32)
+ volumes_16_l = MaxPooling3D(pool_size=(2, 2, 2), name='ref_m1')(rn3)
+
+ print("volumes_16_l_shape", volumes_16_l.shape) # volumes_16_l_shape (?,16,16,16,32)
+
+ rn5 = Conv3D(filters=64, kernel_size=(4, 4, 4), padding='same', data_format="channels_last", name='ref_c2')(
+ volumes_16_l)
+ rn6 = BatchNormalization()(rn5)
+ rn7 = LeakyReLU(alpha=.2)(rn6)
+ print("rn7.shape", rn7.shape) # rn7.shape (?, 16, 16, 16, 64)
+ volumes_8_l = MaxPooling3D(pool_size=(2, 2, 2), name='ref_m2')(rn7)
+
+ print("volumes_8_l_shape", volumes_8_l.shape) # volumes_8_l_shape (?,8,8,8,64)
+
+ rn9 = Conv3D(filters=128, kernel_size=(4, 4, 4), padding='same', data_format="channels_last",
+ name='ref_c3')(volumes_8_l)
+ rn10 = BatchNormalization()(rn9)
+ rn11 = LeakyReLU(alpha=.2)(rn10)
+ print("rn11.shape", rn11.shape) # rn11.shape (?, 8, 8, 8, 128)
+ volumes_4_l = MaxPooling3D(pool_size=(2, 2, 2), name='ref_m3')(rn11)
+
+ print("volumes_4_l_shape", volumes_4_l.shape) # volumes_4_l_shape (?,4,4,4,128)
+
+ flatten_features = tf.reshape(volumes_4_l, [-1, 8192], name="ref_fc1_in")
+ with tf.variable_scope('ref_fc'):
+ fc1 = Dense(units=2048, activation='relu', name='ref_fc1')(flatten_features)
+ # fc1=tanh(fc1)
+
+ fc1 = relu(fc1, alpha=0.0, max_value=None, threshold=0.0)
+
+ print("fc1_shape", fc1.shape) # fc1_shape (?,4,4,4,2048)
+
+ fc2 = Dense(units=8192, activation='relu', name='ref_fc2')(fc1)
+ # fc2=tanh(fc2)
+ fc2 = relu(fc2, alpha=0.0, max_value=None, threshold=0.0)
+
+ print("fc2_shape", fc2.shape) # fc2_shape (?,4,4,4,8192)
+
+ fc2 = tf.reshape(fc2, [-1, 4, 4, 4, 128], name="ref_fc2_out")
+
+ with tf.variable_scope('ref_Dec'):
+ reshaped_1 = Add()([fc2, volumes_4_l])
+
+ print("reshaped_1.shape", reshaped_1.shape) # reshaped_1.shape (?,4,4,4,128)
+
+ rn13 = Conv3DTranspose(filters=64, kernel_size=(4, 4, 4), padding='same', data_format="channels_last",
+ name='ref_d1', strides=(2, 2, 2))(reshaped_1)
+
+ rn14 = BatchNormalization()(rn13)
+ volumes_4_r = relu(rn14, alpha=0.0, max_value=None, threshold=0.0)
+
+ print("volumes_4_r_shape", volumes_4_r.shape) # volumes_4_r_shape (?,8,8,8,64)
+
+ reshaped_2 = Add()([volumes_4_r, volumes_8_l])
+
+ print("reshaped_2_shape", reshaped_2.shape) # volumes_2_shape (?,8,8,8,64)
+
+ rn16 = Conv3DTranspose(filters=32, kernel_size=(4, 4, 4), padding='same', data_format="channels_last",
+ name='ref_d2', strides=(2, 2, 2))(reshaped_2)
+ rn17 = BatchNormalization()(rn16)
+ volumes_8_r = relu(rn17, alpha=0.0, max_value=None, threshold=0.0)
+
+ reshaped_3 = Add()([volumes_8_r, volumes_16_l])
+
+ print("reshaped_3_shape", reshaped_3.shape) # reshaped_3_shape (?,16,16,16,32)
+
+ rn19 = Conv3DTranspose(filters=1, kernel_size=(4, 4, 4), padding='same', data_format="channels_last",
+ name='ref_d3', strides=(2, 2, 2))(volumes_8_r)
+ print("rn19_shape", rn19.shape) # rn19_shape (?, ?, ?, ?, 1)
+ # volumes_16_r= tf.nn.sigmoid(rn19,name='ref_sigmoid1')
+ # reshape_4=volumes_16_r ####################
+
+ reshape_4 = Add()([rn19, input_volumes_32])
+ reshape_4 = (reshape_4 * 0.5)
+ print("reshape_4_5", reshape_4.shape) # reshape_4_5 (?,32,32,32,1)
+
+ reshape_4 = tf.nn.sigmoid(reshape_4, name='ref_sigmoid1')
+
+ print("reshape_4_sig_shape", reshape_4.shape) # reshape_4_sig_shape (?,32,32,32,1)
+
+ reshape_5 = tf.reshape(reshape_4, [-1, vox_res32, vox_res32, vox_res32], name="ref_out")
+
+ return reshape_5
+
+
+def attsets_fc(x, out_ele_num):
+ with tf.variable_scope('att_fc'):
+ in_ele_num = tf.shape(x)[1]
+ in_ele_len = int(x.get_shape()[2])
+ out_ele_len = in_ele_len
+ print("out_ele_len ", out_ele_len)
+ ####################
+ x_1st = x
+ x_1st_tp = tf.reshape(x_1st, [-1, in_ele_len], name="att_in")
+ weights_1st = tools.Ops.fc(x_1st_tp, out_d=out_ele_num * out_ele_len, name="att")
+
+ ########## option 1
+ weights_1st = weights_1st
+ ########## option 2
+ # weights_1st = tf.nn.tanh(weights_1st)
+
+ ####################
+ weights_1st = tf.reshape(weights_1st, [-1, in_ele_num, out_ele_num, out_ele_len], name="att_fc_out")
+ weights_1st = tf.nn.softmax(weights_1st, 1)
+ x_1st = tf.tile(x_1st[:, :, None, :], [1, 1, out_ele_num, 1])
+ x_1st = x_1st * weights_1st
+ x_1st = tf.reduce_sum(x_1st, axis=1)
+ x_1st = tf.reshape(x_1st, [-1, out_ele_num * out_ele_len], name="att_out")
+ return x_1st, weights_1st
+
+
+#####################################
+class Network:
+ def __init__(self):
+ self.train_mod_dir = './Model/train_mod/'
+ self.train_sum_dir = './Model/train_sum/'
+ self.test_res_dir = './Model/test_res/'
+ self.test_sum_dir = './Model/test_sum/'
+
+ print('re_train : ', re_train)
+ if os.path.exists(self.test_res_dir):
+ if re_train:
+ print('test_res_dir and files kept!')
+ else:
+ shutil.rmtree(self.test_res_dir)
+ os.makedirs(self.test_res_dir)
+ print('test_res_dir: deleted and then created!')
+ else:
+ os.makedirs(self.test_res_dir)
+ print('test_res_dir: created!')
+
+ if os.path.exists(self.train_mod_dir):
+ if re_train:
+ if os.path.exists(self.train_mod_dir + 'model.cptk.data-00000-of-00001'):
+ print('model found! will be reused!')
+ else:
+ print('model not found! error!')
+ # exit()
+ else:
+ shutil.rmtree(self.train_mod_dir)
+ os.makedirs(self.train_mod_dir)
+ print('train_mod_dir: deleted and then created!')
+ else:
+ os.makedirs(self.train_mod_dir)
+ print('train_mod_dir: created!')
+
+ if os.path.exists(self.train_sum_dir):
+ if re_train:
+ print('train_sum_dir and files kept!')
+ else:
+ shutil.rmtree(self.train_sum_dir)
+ os.makedirs(self.train_sum_dir)
+ print('train_sum_dir: deleted and then created!')
+ else:
+ os.makedirs(self.train_sum_dir)
+ print('train_sum_dir: created!')
+
+ if os.path.exists(self.test_sum_dir):
+ if re_train:
+ print('test_sum_dir and files kept!')
+ else:
+ shutil.rmtree(self.test_sum_dir)
+ os.makedirs(self.test_sum_dir)
+ print('test_sum_dir: deleted and then created!')
+ else:
+ os.makedirs(self.test_sum_dir)
+ print('test_sum_dir: created!')
+
+ def base_r2n2(self, X_rgb):
+ with tf.variable_scope('Encoder'):
+ im_num = tf.shape(X_rgb)[1]
+
+ [_, _, d1, d2, cc] = X_rgb.get_shape()
+ X_rgb = tf.reshape(X_rgb, [-1, int(d1), int(d2), int(cc)], name="en_in")
+ print("Network Structure")
+ print("base_r2n2", X_rgb.shape) # base_r2n2 (?, 127, 127, 3)
+
+ # plot_buf_1= tf.reshape(X_rgb, [-1, 127, 127, 3]) ###################
+ # tf.summary.image("Input", plot_buf_1)###############################
+
+ en_c = [96, 128, 256, 256, 256, 256]
+ l1 = tools.Ops.xxlu(tools.Ops.conv2d(X_rgb, k=7, out_c=en_c[0], str=1, name='en_c1'), label='lrelu')
+ print("l1_r2n", l1.shape) # l1_r2n (?, 127, 127, 96)
+
+ l2 = tools.Ops.xxlu(tools.Ops.conv2d(l1, k=3, out_c=en_c[0], str=1, name='en_c2'), label='lrelu')
+ l2 = tools.Ops.maxpool2d(l2, k=2, s=2, name='en_mp1')
+ print("l2_r2n", l2.shape) # l2_r2n (?, 64, 64, 96)
+
+ # plot_buf_1= tf.reshape(l2, [-1, 32, 32, 3]) #########################
+ # tf.summary.image("L2_MP_en", plot_buf_1)#############################
+
+ l3 = tools.Ops.xxlu(tools.Ops.conv2d(l2, k=3, out_c=en_c[1], str=1, name='en_c3'), label='lrelu')
+ print("l3_r2n", l3.shape) # l3_r2n (?, 64, 64, 128)
+ l4 = tools.Ops.xxlu(tools.Ops.conv2d(l3, k=3, out_c=en_c[1], str=1, name='en_c4'), label='lrelu')
+ print("l4_r2n", l4.shape) # l4_r2n (?, 64, 64, 128)
+ l22 = tools.Ops.conv2d(l2, k=1, out_c=en_c[1], str=1, name='en_c22')
+ print("l22_r2n", l22.shape) # l22_r2n (?, 64, 64, 128)
+ l4 = l4 + l22
+ l4 = tools.Ops.maxpool2d(l4, k=2, s=2, name='en_mp2')
+ print("l4+l22_r2n", l4.shape) # l4+l22_r2n (?, 32, 32, 128)
+
+ # plot_buf_1= tf.reshape(l4, [-1, 32, 32, 3]) ##########################
+ # tf.summary.image("l4+l22_en", plot_buf_1)#############################
+
+ l5 = tools.Ops.xxlu(tools.Ops.conv2d(l4, k=3, out_c=en_c[2], str=1, name='en_c5'), label='lrelu')
+ print("l5_r2n", l5.shape) # l5_r2n (?, 32, 32, 256)
+ l6 = tools.Ops.xxlu(tools.Ops.conv2d(l5, k=3, out_c=en_c[2], str=1, name='en_c6'), label='lrelu')
+ print("l6_r2n", l6.shape) # l6_r2n (?, 32, 32, 256)
+ l44 = tools.Ops.conv2d(l4, k=1, out_c=en_c[2], str=1, name='en_c44')
+ print("l44_r2n", l44.shape) # l44_r2n (?, 32, 32, 256)
+ l6 = l6 + l44
+ l6 = tools.Ops.maxpool2d(l6, k=2, s=2, name='en_mp3')
+ print("l6+l44_r2n", l6.shape) # l6+l44_r2n (?, 16, 16, 256)
+
+ # plot_buf_1= tf.reshape(l6, [-1, 16, 16, 3]) ##########################
+ # tf.summary.image("l6+l44_en", plot_buf_1)#############################
+
+ l7 = tools.Ops.xxlu(tools.Ops.conv2d(l6, k=3, out_c=en_c[3], str=1, name='en_c7'), label='lrelu')
+ print("l7_r2n", l7.shape) # l7_r2n (?, 16, 16, 256)
+ l8 = tools.Ops.xxlu(tools.Ops.conv2d(l7, k=3, out_c=en_c[3], str=1, name='en_c8'), label='lrelu')
+ print("l8_r2n", l8.shape) # l8_r2n (?, 16, 16, 256)
+ # l66=tools.Ops.conv2d(l6, k=1, out_c=en_c[3], str=1, name='en_c66')
+ # print("l66_r2n",l66.shape)
+ # l8=l8+l66
+ l8 = tools.Ops.maxpool2d(l8, k=2, s=2, name='en_mp4')
+ print("l8_r2n", l8.shape) # l8_r2n (?, 8, 8, 256)
+
+ # plot_buf_1= tf.reshape(l8, [-1, 8, 8, 3]) ########################
+ # tf.summary.image("l8_en", plot_buf_1)#############################
+
+ l9 = tools.Ops.xxlu(tools.Ops.conv2d(l8, k=3, out_c=en_c[4], str=1, name='en_c9'), label='lrelu')
+ print("l9_r2n", l9.shape) # l9_r2n (?, 8, 8, 256)
+ l10 = tools.Ops.xxlu(tools.Ops.conv2d(l9, k=3, out_c=en_c[4], str=1, name='en_c10'), label='lrelu')
+ print("l10_r2n", l10.shape) # l10_r2n (?, 8, 8, 256)
+ l88 = tools.Ops.conv2d(l8, k=1, out_c=en_c[4], str=1, name='en_c88')
+ print("l88_r2n", l88.shape) # l88_r2n (?, 8, 8, 256)
+ l10 = l10 + l88
+ l10 = tools.Ops.maxpool2d(l10, k=2, s=2, name='en_mp5')
+ print("l10_r2n", l10.shape) # l10_r2n (?, 4, 4, 256)
+
+ # plot_buf_1= tf.reshape(l10, [-1, 4, 4, 3]) ########################
+ # tf.summary.image("l10_en", plot_buf_1)#############################
+
+ l11 = tools.Ops.xxlu(tools.Ops.conv2d(l10, k=3, out_c=en_c[5], str=1, name='en_c11'), label='lrelu')
+ print("l11_r2n", l11.shape) # l11_r2n (?, 4, 4, 256)
+ l12 = tools.Ops.xxlu(tools.Ops.conv2d(l11, k=3, out_c=en_c[5], str=1, name='en_c12'), label='lrelu')
+ print("l12_r2n", l12.shape) # l12_r2n (?, 4, 4, 256)
+ l1010 = tools.Ops.conv2d(l10, k=1, out_c=en_c[5], str=1, name='en_c1010')
+ print("l1010_r2n", l1010.shape) # l1010_r2n (?, 4, 4, 256)
+ l12 = l12 + l1010
+ l12 = tools.Ops.maxpool2d(l12, k=2, s=2, name='en_mp6')
+ print("l12_r2n", l12.shape) # l12_r2n (?, 2, 2, 256)
+
+ # plot_buf_1= tf.reshape(l12, [-1, 2, 2, 3]) ########################
+ # tf.summary.image("l12_en", plot_buf_1)#############################
+
+ [_, d1, d2, cc] = l12.get_shape()
+ l12 = tf.reshape(l12, [-1, int(d1) * int(d2) * int(cc)], name="en_fc1_in")
+ print("fc1_input_r2n", l12.shape) # fc1_input_r2n (?, 1024)
+ fc = tools.Ops.xxlu(tools.Ops.fc(l12, out_d=1024, name='en_fc1'), label='lrelu')
+ print("fc1_output_r2n", fc.shape) # fc1_output_r2n (?, 1024)
+
+ with tf.variable_scope('Att_Net'):
+ #### use fc attention
+ input = tf.reshape(fc, [-1, im_num, 1024], name="Att_fc_in")
+ print("att_fc_in_r2n", input.shape) # att_fc_in_r2n (?, ?, 1024)
+ latent_3d, weights = attsets_fc(input, out_ele_num=1)
+ print("att_fc_out_r2n", latent_3d.shape) # att_fc_out_r2n (?, 1024)
+
+ with tf.variable_scope('Decoder'):
+ ####
+ latent_3d = tools.Ops.xxlu(tools.Ops.fc(latent_3d, out_d=4 * 4 * 4 * 128, name='de_fc2'), label='lrelu')
+ print("fc3_out_r2n", latent_3d.shape) # fc3_out_r2n (?, 8192)
+ latent_3d = tf.reshape(latent_3d, [-1, 4, 4, 4, 128], name="de_fc2_out")
+
+ ####
+
+ de_c = [128, 128, 128, 64, 32, 1]
+
+ print("d1_in_r2n", latent_3d.shape) # d1_in_r2n (?, 4, 4, 4, 128)
+ d1 = tools.Ops.xxlu(tools.Ops.deconv3d(latent_3d, k=3, out_c=de_c[1], str=2, name='de_c1'), label='lrelu')
+ print("d1_out_r2n", d1.shape) # d1_out_r2n (?, 8, 8, 8, 128)
+ d2 = tools.Ops.xxlu(tools.Ops.deconv3d(d1, k=3, out_c=de_c[1], str=1, name='de_c2'), label='lrelu')
+ print("d2_out_r2n", d2.shape) # d2_out_r2n (?, 8, 8, 8, 128)
+ d00 = tools.Ops.deconv3d(latent_3d, k=1, out_c=de_c[1], str=2, name='de_c00')
+ print("d00_out_r2n", d00.shape) # d00_out_r2n (?, 8, 8, 8, 128)
+ d2 = d2 + d00
+ print("d2+d00_out_r2n", d2.shape) # d2+d00_out_r2n (?, 8, 8, 8, 128)
+
+ # plot_buf_1= tf.reshape(d2, [-1, 8, 8, 4]) ################################
+ # tf.summary.image("d2+d00_out_de", plot_buf_1)#############################
+
+ d3 = tools.Ops.xxlu(tools.Ops.deconv3d(d2, k=3, out_c=de_c[2], str=2, name='de_c3'), label='lrelu')
+ print("d3_out_r2n", d3.shape) # d3_out_r2n (?, 16, 16, 16, 128)
+ d4 = tools.Ops.xxlu(tools.Ops.deconv3d(d3, k=3, out_c=de_c[2], str=1, name='de_c4'), label='lrelu')
+ print("d4_out_r2n", d4.shape) # d4_out_r2n (?, 16, 16, 16, 128)
+ d22 = tools.Ops.deconv3d(d2, k=1, out_c=de_c[2], str=2, name='de_c22')
+ print("d22_out_r2n", d22.shape) # d22_out_r2n (?, 16, 16, 16, 128)
+ d4 = d4 + d22
+ print("d4+d22_out_r2n", d4.shape) # d4+d22_out_r2n (?, 16, 16, 16, 128)
+
+ # plot_buf_1= tf.reshape(d4, [-1, 16, 16, 4]) ##############################
+ # tf.summary.image("d4+d22_out_de", plot_buf_1)#############################
+
+ d5 = tools.Ops.xxlu(tools.Ops.deconv3d(d4, k=3, out_c=de_c[3], str=2, name='de_c5'), label='lrelu')
+ print("d5_out_r2n", d5.shape) # d5_out_r2n (?, 32, 32, 32, 64)
+ d6 = tools.Ops.xxlu(tools.Ops.deconv3d(d5, k=3, out_c=de_c[3], str=1, name='de_c6'), label='lrelu')
+ print("d6_out_r2n", d6.shape) # d6_out_r2n (?, 32, 32, 32, 64)
+ d44 = tools.Ops.deconv3d(d4, k=1, out_c=de_c[3], str=2, name='de_c44')
+ print("d44_out_r2n", d44.shape) # d44_out_r2n (?, 32, 32, 32, 64)
+ d6 = d6 + d44
+ print("d6+d44_out_r2n", d6.shape) # d6+d44_out_r2n (?, 32, 32, 32, 64)
+
+ # plot_buf_1= tf.reshape(d6, [-1, 32, 32, 4]) ##############################
+ # tf.summary.image("d6+d44_out_de", plot_buf_1)#############################
+
+ d7 = tools.Ops.xxlu(tools.Ops.deconv3d(d6, k=3, out_c=de_c[4], str=1, name='de_c7'), label='lrelu')
+ print("d7_out_r2n", d7.shape) # d7_out_r2n (?, 32, 32, 32, 32)
+ d8 = tools.Ops.xxlu(tools.Ops.deconv3d(d7, k=3, out_c=de_c[4], str=1, name='de_c8'), label='lrelu')
+ print("d8_out_r2n", d8.shape) # d8_out_r2n (?, 32, 32, 32, 32)
+ d77 = tools.Ops.xxlu(tools.Ops.deconv3d(d7, k=3, out_c=de_c[4], str=1, name='de_c77'), label='lrelu')
+ print("d77_out_r2n", d77.shape) # d77_out_r2n (?, 32, 32, 32, 32)
+ d8 = d8 + d77
+ print("d8+d77_out_r2n", d8.shape) # d8+d77_out_r2n (?, 32, 32, 32, 32)
+
+ # plot_buf_1= tf.reshape(d8, [-1, 32, 32, 4]) ##############################
+ # tf.summary.image("d8+d77_out_de", plot_buf_1)#############################
+
+ d11 = tools.Ops.deconv3d(d8, k=3, out_c=de_c[5], str=1, name='de_c9')
+ print("d11_out_r2n", d11.shape) # d11_out_r2n (?, 32, 32, 32, 1)
+
+ # plot_buf_1= tf.reshape(d11, [-1, 32, 32,4]) ##########################
+ # tf.summary.image("Ref_input", plot_buf_1)#############################
+
+ ref_in = tf.reshape(d11, [-1, vox_res32, vox_res32, vox_res32], name="ref_in") ###
+
+ y = tf.nn.sigmoid(d11, name='de_sigmoid')
+
+ att_o = tf.reshape(y, [-1, vox_res32, vox_res32, vox_res32], name="de_out")
+
+ print("att_out_shape", att_o.shape) # att_out_shape (?, 32, 32, 32)
+
+ with tf.variable_scope('ref_net'):
+ ref_o = refiner_network(ref_in)
+
+ # plot_buf_1= tf.reshape(ref_o, [-1, 32, 32,4]) ######################
+ # tf.summary.image("Ref_Out", plot_buf_1)#############################
+
+ return ref_o, att_o, weights
+
+ def build_graph(self):
+ img_res = 127
+ vox_res = 32
+ self.X_rgb = tf.placeholder(shape=[None, None, img_res, img_res, 3], dtype=tf.float32)
+ self.Y_vox = tf.placeholder(shape=[None, vox_res, vox_res, vox_res], dtype=tf.float32)
+ self.lr = tf.placeholder(tf.float32)
+ self.refine_lr = tf.placeholder(tf.float32)
+ with tf.device('/cpu:0'):
+ self.Y_pred, self.vae_o, self.weights = self.base_r2n2(self.X_rgb)
+ tf.summary.histogram('Attsets_Weights', self.weights)
+ with tf.device('/cpu:0'):
+ ### rec loss
+ print("reached")
+ with tf.variable_scope('Loss_Fun'):
+ Y_vox_ = tf.reshape(self.Y_vox, shape=[-1, vox_res ** 3])
+ Y_pred_ = tf.reshape(self.Y_pred, shape=[-1, vox_res ** 3])
+ vae_o_ = tf.reshape(self.vae_o, shape=[-1, vox_res ** 3])
+
+ self.vae_loss = tf.reduce_mean(-tf.reduce_mean(Y_vox_ * tf.log(vae_o_ + 1e-8), reduction_indices=[1]) -
+ tf.reduce_mean((1 - Y_vox_) * tf.log(1 - vae_o_ + 1e-8),
+ reduction_indices=[1]))
+ self.rec_loss = tf.reduce_mean(-tf.reduce_mean(Y_vox_ * tf.log(Y_pred_ + 1e-8), reduction_indices=[1]) -
+ tf.reduce_mean((1 - Y_vox_) * tf.log(1 - Y_pred_ + 1e-8),
+ reduction_indices=[1]))
+ sum_rec_loss = tf.summary.scalar('rec_loss', self.rec_loss)
+ self.sum_merged = sum_rec_loss
+ tf.summary.histogram('rec_loss', self.rec_loss)
+ tf.summary.scalar("vae-loss", self.vae_loss)
+ tf.summary.histogram("vae_loss", self.vae_loss)
+ self.mean_loss = tf.div(x=tf.math.add(x=self.vae_loss, y=self.rec_loss, name='add_loss'), y=2,
+ name='mean_loss')
+ tf.summary.histogram("mean_vae_loss", self.mean_loss)
+ tf.summary.scalar("mean_vae_loss", self.mean_loss)
+
+ with tf.variable_scope('Evaluation_Metric'):
+ gt_vox = Y_vox_
+ self.iou_ref = metric_iou(Y_pred_, gt_vox)
+ tf.summary.scalar('iou_refiner', self.iou_ref)
+ tf.summary.histogram('iou_refiner', self.iou_ref)
+ self.iou_vae = metric_iou(vae_o_, gt_vox)
+ tf.summary.scalar('iou_vae', self.iou_vae)
+ tf.summary.histogram("iou_vae", self.iou_vae)
+
+ with tf.variable_scope('Optimization'):
+ base_en_var = [var for var in tf.trainable_variables() if var.name.startswith('Encoder/en')]
+ base_dec_var = [var for var in tf.trainable_variables() if var.name.startswith('Decoder/de')]
+ att_var = [var for var in tf.trainable_variables() if var.name.startswith('Att_Net/att')]
+ refine_var = [var for var in tf.trainable_variables() if var.name.startswith('ref_net/ref')]
+ # self.refine_optim = tf.train.AdamOptimizer(learning_rate=self.refine_lr).minimize(self.rec_loss, var_list=refine_var)
+ # self.base_en_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=base_en_var)
+ # self.base_de_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=base_dec_var)
+ # self.att_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr).minimize(self.vae_loss, var_list=att_var)
+ self.refine_optim = tf.train.AdamOptimizer(learning_rate=self.refine_lr)
+ self.base_en_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr)
+ self.base_de_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr)
+ self.att_optim2 = tf.train.AdamOptimizer(learning_rate=self.lr)
+
+ '''
+
+ loss_scale_manager_1 = ExponentialUpdateLossScaleManager(init_loss_scale=2 ** 32, incr_every_n_steps=1000,
+ decr_every_n_nan_or_inf=2, decr_ratio=0.5)
+ loss_scale_manager_2 = ExponentialUpdateLossScaleManager(init_loss_scale=2 ** 32, incr_every_n_steps=1000,
+ decr_every_n_nan_or_inf=2, decr_ratio=0.5)
+ loss_scale_manager_3 = ExponentialUpdateLossScaleManager(init_loss_scale=2 ** 32, incr_every_n_steps=1000,
+ decr_every_n_nan_or_inf=2, decr_ratio=0.5)
+ loss_scale_manager_4 = ExponentialUpdateLossScaleManager(init_loss_scale=2 ** 32, incr_every_n_steps=1000,
+ decr_every_n_nan_or_inf=2, decr_ratio=0.5)
+ '''
+ loss_scale_manager_1 = FixedLossScaleManager(loss_scale=8589935000.0, enable_overflow_check=False)
+ loss_scale_manager_2 = FixedLossScaleManager(loss_scale=2147483600.0, enable_overflow_check=False)
+ loss_scale_manager_3 = FixedLossScaleManager(loss_scale=2147483600.0, enable_overflow_check=False)
+ loss_scale_manager_4 = FixedLossScaleManager(loss_scale=4294967300.0, enable_overflow_check=False)
+
+ self.refine_optim = NPULossScaleOptimizer(self.refine_optim, loss_scale_manager_1).minimize(self.rec_loss,
+ var_list=refine_var)
+ self.base_en_optim2 = NPULossScaleOptimizer(self.base_en_optim2, loss_scale_manager_2).minimize(
+ self.vae_loss, var_list=base_en_var)
+ self.base_de_optim2 = NPULossScaleOptimizer(self.base_de_optim2, loss_scale_manager_3).minimize(
+ self.vae_loss, var_list=base_dec_var)
+ self.att_optim2 = NPULossScaleOptimizer(self.att_optim2, loss_scale_manager_4).minimize(self.vae_loss,
+ var_list=att_var)
+
+ print("total weights:", tools.Ops.variable_count())
+ self.saver = tf.train.Saver(max_to_keep=1)
+
+ config = tf.ConfigProto(allow_soft_placement=True)
+ config.gpu_options.allow_growth = True
+ config.gpu_options.visible_device_list = '0,1'
+
+ self.sess = tf.Session(config=npu_config_proto(config_proto=config))
+ #config = npu_config_proto(config_proto=config)
+ self.merged = tf.summary.merge_all()
+ self.sum_writer_train = tf.summary.FileWriter(self.train_sum_dir, self.sess.graph)
+ self.sum_writer_test = tf.summary.FileWriter(self.test_sum_dir, self.sess.graph)
+
+ #######################
+ path = self.train_mod_dir
+ # path = './Model_released/' # retrain the released model
+
+ if os.path.isfile(path + 'model.cptk.data-00000-of-00001'):
+ print("restoring saved model!")
+ self.saver.restore(self.sess, path + 'model.cptk')
+ else:
+ self.sess.run(tf.global_variables_initializer())
+
+ return 0
+
+ def train(self, data):
+ for epoch in range(0, 50, 1):
+ train_view_num = 24 ##!!!!!!!!!!!
+ data.shuffle_train_files(epoch, train_mv=train_view_num)
+ total_train_batch_num = data.total_train_batch_num # int(len(self.X_rgb_train_files)/(self.batch_size*train_mv))
+ print('total_train_batch_num:', total_train_batch_num)
+ for i in range(total_train_batch_num):
+ #### training
+ X_rgb_bat, Y_vox_bat = data.load_X_Y_train_next_batch(train_mv=train_view_num)
+ print("multi_view_train_X_rgb_bat : ",
+ X_rgb_bat.shape) # np.asarray(X.append(X_rgb[b*train_mv:(b+1)*train_mv,:]))
+
+ print(time.ctime())
+
+ ##### option 1: seperate train, seperate optimize
+ # if epoch<=30:
+ # single_view_train=True
+ # multi_view_train=False
+ # else:
+ # single_view_train=False
+ # multi_view_train=True
+
+ ##### optiion 2: joint train, seperate optimize
+ single_view_train = True
+ multi_view_train = False
+
+ if epoch <= 5:
+ att_lr = .0002
+ ref_lr = .0002
+ if epoch > 5 and epoch <= 40:
+ att_lr = .0001
+ ref_lr = .0001
+ if epoch > 40:
+ att_lr = .00005
+ ref_lr = .00005
+ ########### single view train
+ if single_view_train:
+ rgb = np.reshape(X_rgb_bat, [batch_size * train_view_num, 1, 127, 127, 3])
+ print("single_view_train_rgb_input_shape ", rgb.shape)
+ vox = np.tile(Y_vox_bat[:, None, :, :, :], [1, train_view_num, 1, 1, 1])
+ vox = np.reshape(vox, [batch_size * train_view_num, 32, 32, 32])
+ # vae_loss_c,eee,ddd, rec_loss_c, sum_train,rrr,mean_vae,iou_ref_,iou_vae_ = self.sess.run([self.vae_loss,self.base_en_optim2,self.base_de_optim2,self.rec_loss,self.merged,self.refine_optim,self.mean_loss,self.iou_ref,self.iou_vae],feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr,self.refine_lr: ref_lr})
+
+ vae_loss_c = self.sess.run([self.vae_loss], feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr,
+ self.refine_lr: ref_lr})
+ eee = self.sess.run([self.base_en_optim2],
+ feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr, self.refine_lr: ref_lr})
+ ddd = self.sess.run([self.base_de_optim2],
+ feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr, self.refine_lr: ref_lr})
+ rec_loss_c = self.sess.run([self.rec_loss], feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr,
+ self.refine_lr: ref_lr})
+ sum_train = self.sess.run(self.merged,
+ feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr, self.refine_lr: ref_lr})
+ rrr = self.sess.run([self.refine_optim],
+ feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr, self.refine_lr: ref_lr})
+ mean_vae = self.sess.run([self.mean_loss],
+ feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr, self.refine_lr: ref_lr})
+ iou_ref_ = self.sess.run([self.iou_ref],
+ feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr, self.refine_lr: ref_lr})
+ iou_vae_ = self.sess.run([self.iou_vae], feed_dict={self.X_rgb: rgb, self.Y_vox: vox, self.lr: att_lr,
+ self.refine_lr: ref_lr})
+
+ print('ep:', epoch, 'i:', i, 'train single rec loss:', rec_loss_c)
+ print('ep:', epoch, 'i:', i, 'train single vae loss:', vae_loss_c)
+ print('ep:', epoch, 'i:', i, 'train single mean_vae loss:', mean_vae)
+ print('ep:', epoch, 'i:', i, 'train single ref_iou:', iou_ref_)
+ print('ep:', epoch, 'i:', i, 'train single vae_iou:', iou_vae_)
+
+ ########## multi view train
+
+ if multi_view_train:
+ vae_loss_c, rec_loss_c, _, sum_train, xxx, mean_vae, iou_ref_, iou_vae_ = self.sess.run(
+ [self.vae_loss, self.rec_loss, self.att_optim2, self.merged, self.refine_optim, self.mean_loss,
+ self.iou_ref, self.iou_vae],
+ feed_dict={self.X_rgb: X_rgb_bat, self.Y_vox: Y_vox_bat, self.lr: att_lr, self.refine_lr: ref_lr})
+ print('ep:', epoch, 'i:', i, 'train multi rec loss:', rec_loss_c)
+ print('ep:', epoch, 'i:', i, 'train multi vae loss:', vae_loss_c)
+ print('ep:', epoch, 'i', i, 'train multi mean_vae loss:', mean_vae)
+ print('ep:', epoch, 'i:', i, 'train multi ref_iou:', iou_ref_)
+ print('ep:', epoch, 'i:', i, 'train multi vae_iou:', iou_vae_)
+
+ ############
+ if i % 10 == 0:
+ self.sum_writer_train.add_summary(sum_train, epoch * total_train_batch_num + i)
+
+
+ #### testing
+ if i % 20 == 0:
+ X_rgb_batch, Y_vox_batch = data.load_X_Y_test_next_batch(test_mv=3)
+
+ # vae_pred = tf.get_default_graph().get_tensor_by_name("Decoder/de_out:0")
+ # ref_pred = tf.get_default_graph().get_tensor_by_name("ref_net/ref_Dec/ref_out:0")
+ # gt_vox=Y_vox_batch.astype(np.float32)
+
+ # iou_pred = metric_iou(ref_pred,gt_vox)
+ # tf.summary.scalar("iou",iou_pred)
+
+ rrrr, aaaa, rec_loss_te, qwerty, Y_vox_test_pred, att_pred, sum_test, mean_vae, iou_ref_, iou_vae_ = \
+ self.sess.run([self.refine_optim, self.att_optim2, self.rec_loss, self.vae_loss, self.Y_pred, self.weights,
+ self.merged, self.mean_loss, self.iou_ref, self.iou_vae],
+ feed_dict={self.X_rgb: X_rgb_batch, self.Y_vox: Y_vox_batch, self.lr: 0, self.refine_lr: 0})
+
+ X_rgb_batch = X_rgb_batch.astype(np.float16)
+
+ Y_vox_batch = Y_vox_batch.astype(np.float16)
+ Y_vox_test_pred = Y_vox_test_pred.astype(np.float16)
+ att_pred = att_pred.astype(np.float16)
+ to_save = {'X_test': X_rgb_batch, 'Y_test_pred': Y_vox_test_pred, 'att_pred': att_pred,
+ 'Y_test_true': Y_vox_batch}
+ scipy.io.savemat(self.test_res_dir + 'X_Y_pred_' + str(epoch).zfill(2) + '_' + str(i).zfill(5) + '.mat',
+ to_save, do_compression=True)
+
+ self.sum_writer_test.add_summary(sum_test, epoch * total_train_batch_num + i)
+
+ # iou_ref=evaluate_voxel_prediction(ref_pred_,gt_vox)
+ # iou_vae=evaluate_voxel_prediction(vae_pred_,gt_vox)
+
+ # print("Ref_iou:",iou_ref)
+ # print("Vae_iou:",iou_vae)
+
+ # plot_list_iou.append(iou_ref)
+ # plot_list_i.append((i/50))
+ # graph_plot(plot_list_iou,plot_list_i)
+ print('ep:', epoch, 'i:', i, 'test rec loss:', rec_loss_te)
+ print('ep:', epoch, 'i:', i, 'test vae loss:', qwerty)
+ print('ep:', epoch, 'i:', i, 'test mean_vae loss:', mean_vae)
+ print('ep:', epoch, 'i:', i, 'test ref_iou:', iou_ref_)
+ print('ep:', epoch, 'i:', i, 'test vae_iou:', iou_vae_)
+
+ #### model saving
+ if i % 100 == 0:
+ self.saver.save(self.sess, save_path=self.train_mod_dir + '/' + str(epoch) + str(i) + '/' + 'model.cptk')
+ print('epoch:', epoch, 'i:', i, 'model saved!')
+
+
+# plt.show()
+
+
+##########
+if __name__ == '__main__':
+ npu_keras_sess = set_keras_session_npu_config()
+
+ net = Network() # net=object to create instance
+
+ print("network compleated") ###
+
+ net.build_graph()
+ print("graph compleated")
+
+ data = tools.Data(config)
+ print("tools.data compleated")
+
+ print('training data')
+
+ net.train(data)
+
+ close_session(npu_keras_sess)
--
Gitee
From 218fc5953e1b38102262f45cdb8c41fce154302c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:46:25 +0000
Subject: [PATCH 09/16] =?UTF-8?q?=E5=8E=BB=E6=8E=89code=E7=9B=AE=E5=BD=95?=
=?UTF-8?q?=E5=B1=82=E7=BA=A7=20=E5=A2=9E=E5=8A=A0=E6=96=87=E4=BB=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../binvox_rw.py | 284 ++++++++++++++++++
.../demo_AttSets.py | 186 ++++++++++++
.../export_obj.py | 45 +++
3 files changed, 515 insertions(+)
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/binvox_rw.py
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/demo_AttSets.py
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/export_obj.py
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/binvox_rw.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/binvox_rw.py
new file mode 100644
index 000000000..40005d24a
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/binvox_rw.py
@@ -0,0 +1,284 @@
+# Copyright (C) 2012 Daniel Maturana
+# This file is part of binvox-rw-py.
+#
+# binvox-rw-py is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# binvox-rw-py is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+#
+# ============================================================================
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# along with binvox-rw-py. If not, see .
+#
+
+"""
+Binvox to Numpy and back.
+>>> import numpy as np
+>>> import binvox_rw
+>>> with open('chair.binvox', 'rb') as f:
+... m1 = binvox_rw.read_as_3d_array(f)
+...
+>>> m1.dims
+[32, 32, 32]
+>>> m1.scale
+41.133000000000003
+>>> m1.translate
+[0.0, 0.0, 0.0]
+>>> with open('chair_out.binvox', 'wb') as f:
+... m1.write(f)
+...
+>>> with open('chair_out.binvox', 'rb') as f:
+... m2 = binvox_rw.read_as_3d_array(f)
+...
+>>> m1.dims==m2.dims
+True
+>>> m1.scale==m2.scale
+True
+>>> m1.translate==m2.translate
+True
+>>> np.all(m1.data==m2.data)
+True
+>>> with open('chair.binvox', 'rb') as f:
+... md = binvox_rw.read_as_3d_array(f)
+...
+>>> with open('chair.binvox', 'rb') as f:
+... ms = binvox_rw.read_as_coord_array(f)
+...
+>>> data_ds = binvox_rw.dense_to_sparse(md.data)
+>>> data_sd = binvox_rw.sparse_to_dense(ms.data, 32)
+>>> np.all(data_sd==md.data)
+True
+>>> # the ordering of elements returned by numpy.nonzero changes with axis
+>>> # ordering, so to compare for equality we first lexically sort the voxels.
+>>> np.all(ms.data[:, np.lexsort(ms.data)] == data_ds[:, np.lexsort(data_ds)])
+True
+"""
+from npu_bridge.npu_init import *
+
+import numpy as np
+
+class Voxels(object):
+ """ Holds a binvox model.
+ data is either a three-dimensional numpy boolean array (dense representation)
+ or a two-dimensional numpy float array (coordinate representation).
+ dims, translate and scale are the model metadata.
+ dims are the voxel dimensions, e.g. [32, 32, 32] for a 32x32x32 model.
+ scale and translate relate the voxels to the original model coordinates.
+ To translate voxel coordinates i, j, k to original coordinates x, y, z:
+ x_n = (i+.5)/dims[0]
+ y_n = (j+.5)/dims[1]
+ z_n = (k+.5)/dims[2]
+ x = scale*x_n + translate[0]
+ y = scale*y_n + translate[1]
+ z = scale*z_n + translate[2]
+ """
+
+ def __init__(self, data, dims, translate, scale, axis_order):
+ self.data = data
+ self.dims = dims
+ self.translate = translate
+ self.scale = scale
+ assert (axis_order in ('xzy', 'xyz'))
+ self.axis_order = axis_order
+
+ def clone(self):
+ data = self.data.copy()
+ dims = self.dims[:]
+ translate = self.translate[:]
+ return Voxels(data, dims, translate, self.scale, self.axis_order)
+
+ def write(self, fp):
+ write(self, fp)
+
+def read_header(fp):
+ """ Read binvox header. Mostly meant for internal use.
+ """
+ line = fp.readline().strip()
+ if not line.startswith(b'#binvox'):
+ raise IOError('Not a binvox file')
+ dims = list(map(int, fp.readline().strip().split(b' ')[1:]))
+ translate = list(map(float, fp.readline().strip().split(b' ')[1:]))
+ scale = list(map(float, fp.readline().strip().split(b' ')[1:]))[0]
+ line = fp.readline()
+ return dims, translate, scale
+
+def read_as_3d_array(fp, fix_coords=True):
+ """ Read binary binvox format as array.
+ Returns the model with accompanying metadata.
+ Voxels are stored in a three-dimensional numpy array, which is simple and
+ direct, but may use a lot of memory for large models. (Storage requirements
+ are 8*(d^3) bytes, where d is the dimensions of the binvox model. Numpy
+ boolean arrays use a byte per element).
+ Doesn't do any checks on input except for the '#binvox' line.
+ """
+ dims, translate, scale = read_header(fp)
+ raw_data = np.frombuffer(fp.read(), dtype=np.uint8)
+ # if just using reshape() on the raw data:
+ # indexing the array as array[i,j,k], the indices map into the
+ # coords as:
+ # i -> x
+ # j -> z
+ # k -> y
+ # if fix_coords is true, then data is rearranged so that
+ # mapping is
+ # i -> x
+ # j -> y
+ # k -> z
+ values, counts = raw_data[::2], raw_data[1::2]
+ data = np.repeat(values, counts).astype(np.bool)
+ data = data.reshape(dims)
+ if fix_coords:
+ # xzy to xyz TODO the right thing
+ data = np.transpose(data, (0, 2, 1))
+ axis_order = 'xyz'
+ else:
+ axis_order = 'xzy'
+ return Voxels(data, dims, translate, scale, axis_order)
+
+def read_as_coord_array(fp, fix_coords=True):
+ """ Read binary binvox format as coordinates.
+ Returns binvox model with voxels in a "coordinate" representation, i.e. an
+ 3 x N array where N is the number of nonzero voxels. Each column
+ corresponds to a nonzero voxel and the 3 rows are the (x, z, y) coordinates
+ of the voxel. (The odd ordering is due to the way binvox format lays out
+ data). Note that coordinates refer to the binvox voxels, without any
+ scaling or translation.
+ Use this to save memory if your model is very sparse (mostly empty).
+ Doesn't do any checks on input except for the '#binvox' line.
+ """
+ dims, translate, scale = read_header(fp)
+ raw_data = np.frombuffer(fp.read(), dtype=np.uint8)
+
+ values, counts = raw_data[::2], raw_data[1::2]
+
+ sz = np.prod(dims)
+ index, end_index = 0, 0
+ end_indices = np.cumsum(counts)
+ indices = np.concatenate(([0], end_indices[:-1])).astype(end_indices.dtype)
+
+ values = values.astype(np.bool)
+ indices = indices[values]
+ end_indices = end_indices[values]
+
+ nz_voxels = []
+ for index, end_index in zip(indices, end_indices):
+ nz_voxels.extend(range(index, end_index))
+ nz_voxels = np.array(nz_voxels)
+ # TODO are these dims correct?
+ # according to docs,
+ # index = x * wxh + z * width + y; // wxh = width * height = d * d
+
+ x = nz_voxels / (dims[0]*dims[1])
+ zwpy = nz_voxels % (dims[0]*dims[1]) # z*w + y
+ z = zwpy / dims[0]
+ y = zwpy % dims[0]
+ if fix_coords:
+ data = np.vstack((x, y, z))
+ axis_order = 'xyz'
+ else:
+ data = np.vstack((x, z, y))
+ axis_order = 'xzy'
+
+ #return Voxels(data, dims, translate, scale, axis_order)
+ return Voxels(np.ascontiguousarray(data), dims, translate, scale, axis_order)
+
+def dense_to_sparse(voxel_data, dtype=np.int):
+ """ From dense representation to sparse (coordinate) representation.
+ No coordinate reordering.
+ """
+ if voxel_data.ndim!=3:
+ raise ValueError('voxel_data is wrong shape; should be 3D array.')
+ return np.asarray(np.nonzero(voxel_data), dtype)
+
+def sparse_to_dense(voxel_data, dims, dtype=np.bool):
+ if voxel_data.ndim!=2 or voxel_data.shape[0]!=3:
+ raise ValueError('voxel_data is wrong shape; should be 3xN array.')
+ if np.isscalar(dims):
+ dims = [dims]*3
+ dims = np.atleast_2d(dims).T
+ # truncate to integers
+ xyz = voxel_data.astype(np.int)
+ # discard voxels that fall outside dims
+ valid_ix = ~np.any((xyz < 0) | (xyz >= dims), 0)
+ xyz = xyz[:,valid_ix]
+ out = np.zeros(dims.flatten(), dtype=dtype)
+ out[tuple(xyz)] = True
+ return out
+
+#def get_linear_index(x, y, z, dims):
+ #""" Assuming xzy order. (y increasing fastest.
+ #TODO ensure this is right when dims are not all same
+ #"""
+ #return x*(dims[1]*dims[2]) + z*dims[1] + y
+
+def write(voxel_model, fp):
+ """ Write binary binvox format.
+ Note that when saving a model in sparse (coordinate) format, it is first
+ converted to dense format.
+ Doesn't check if the model is 'sane'.
+ """
+ if voxel_model.data.ndim==2:
+ # TODO avoid conversion to dense
+ dense_voxel_data = sparse_to_dense(voxel_model.data, voxel_model.dims)
+ else:
+ dense_voxel_data = voxel_model.data
+
+ fp.write('#binvox 1\n')
+ fp.write('dim '+' '.join(map(str, voxel_model.dims))+'\n')
+ fp.write('translate '+' '.join(map(str, voxel_model.translate))+'\n')
+ fp.write('scale '+str(voxel_model.scale)+'\n')
+ fp.write('data\n')
+ if not voxel_model.axis_order in ('xzy', 'xyz'):
+ raise ValueError('Unsupported voxel model axis order')
+
+ if voxel_model.axis_order=='xzy':
+ voxels_flat = dense_voxel_data.flatten()
+ elif voxel_model.axis_order=='xyz':
+ voxels_flat = np.transpose(dense_voxel_data, (0, 2, 1)).flatten()
+
+ # keep a sort of state machine for writing run length encoding
+ state = voxels_flat[0]
+ ctr = 0
+ for c in voxels_flat:
+ if c==state:
+ ctr += 1
+ # if ctr hits max, dump
+ if ctr==255:
+ fp.write(chr(state))
+ fp.write(chr(ctr))
+ ctr = 0
+ else:
+ # if switch state, dump
+ fp.write(chr(state))
+ fp.write(chr(ctr))
+ state = c
+ ctr = 1
+ # flush out remainders
+ if ctr > 0:
+ fp.write(chr(state))
+ fp.write(chr(ctr))
+
+if __name__ == '__main__':
+ import doctest
+ doctest.testmod()
+
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/demo_AttSets.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/demo_AttSets.py
new file mode 100644
index 000000000..aad6f0277
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/demo_AttSets.py
@@ -0,0 +1,186 @@
+#
+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+from npu_bridge.npu_init import *
+import tensorflow as tf
+import os
+import sys
+sys.path.append('..')
+import tools as tools
+import numpy as np
+import matplotlib.pyplot as plt
+
+GPU='0'
+
+
+def load_real_rgbs(test_mv=5):
+ obj_rgbs_folder ='/home/ma-user/work/3DOR/Data/amazon_real_rgbs/airfilter/'
+ rgbs = []
+ rgbs_views = sorted(os.listdir(obj_rgbs_folder))
+ for v in rgbs_views:
+ if not v.endswith('png'): continue
+
+ rgbs.append(tools.Data.load_single_X_rgb_r2n2(obj_rgbs_folder + v, train=False))
+
+ rgbs = np.asarray(rgbs)
+ x_sample = rgbs[0:test_mv, :, :, :].reshape(1, test_mv, 127, 127, 3)
+ return x_sample, None
+
+def load_shapenet_rgbs(test_mv=8):
+ obj_rgbs_folder = './Data_sample/ShapeNetRendering/03001627/1a6f615e8b1b5ae4dbbc9440457e303e/rendering/'
+ obj_gt_vox_path ='./Data_sample/ShapeNetVox32/03001627/1a6f615e8b1b5ae4dbbc9440457e303e/model.binvox'
+ rgbs=[]
+ rgbs_views = sorted(os.listdir(obj_rgbs_folder))
+ for v in rgbs_views:
+ if not v.endswith('png'): continue
+ rgbs.append(tools.Data.load_single_X_rgb_r2n2(obj_rgbs_folder + v, train=False))
+ rgbs = np.asarray(rgbs)
+ x_sample = rgbs[0:test_mv, :, :, :].reshape(1, test_mv, 127, 127, 3)
+ y_true = tools.Data.load_single_Y_vox(obj_gt_vox_path)
+ return x_sample, y_true
+
+def ttest_demo():
+# model_path = './Model_released/'
+ model_path='/home/ma-user/work/3DOR/Model/train_mod/'
+ if not os.path.isfile(model_path + 'model.cptk.data-00000-of-00001'):
+ print ('please download our released model first!')
+ return
+
+ config = tf.ConfigProto(allow_soft_placement=True)
+ config.gpu_options.visible_device_list = GPU
+
+
+
+ with tf.Session(config=npu_config_proto(config_proto=config)) as sess:
+ saver = tf.train.import_meta_graph(model_path + 'model.cptk.meta', clear_devices=True)
+ saver.restore(sess, model_path + 'model.cptk')
+ print ('model restored!')
+
+ # graph = tf.get_default_graph()
+ # print(graph.get_operations())
+
+ X = tf.get_default_graph().get_tensor_by_name("Placeholder:0")
+ Y_pred = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_9:0")
+
+ plot_data_8 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_8:0")
+ plot_data_7 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_7:0") #############(1,1024)
+ plot_data_6 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_6:0") #############(1,1024)
+ plot_data_5 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_5:0")
+ plot_data_4 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_4:0")
+ plot_data_3 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_3:0")
+ plot_data_2 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_2:0")
+ plot_data_1 = tf.get_default_graph().get_tensor_by_name("r2n/Reshape_1:0")
+
+
+# print("X: ", X.shape) #Tensor("Placeholder:0", shape=(?, ?, 127, 127, 3), dtype=float32)
+# print(Y_pred) #Tensor("r2n/Reshape_9:0", shape=(?, 32, 32, 32), dtype=float32)
+
+# print("x_sample: ", x_sample.shape)
+# print("x_sample_data: ", type(x_sample[:,:,:,:,1]))
+# print(y_pred.shape) ###############################(1, 32, 32, 32) ##############################
+
+
+# x_sample, gt_vox = load_shapenet_rgbs()
+ x_sample, gt_vox = load_real_rgbs()
+
+
+ plot_buf_1= tf.reshape(plot_data_1, [-1, 32, 32, 1])
+ plot_buf_2= tf.reshape(plot_data_2, [-1, 32, 32, 1])
+ plot_buf_3= tf.reshape(plot_data_3, [-1, 32, 32, 1])
+ plot_buf_4= tf.reshape(plot_data_4, [-1, 32, 32, 1])
+ plot_buf_5= tf.reshape(plot_data_5, [-1, 32, 32, 1])
+ plot_buf_6= tf.reshape(plot_data_6, [-1, 32, 32, 1])
+ plot_buf_7= tf.reshape(plot_data_7, [-1, 32, 32, 1])
+ plot_buf_8= tf.reshape(plot_data_8, [-1, 32, 32, 1])
+
+
+# tf.summary.image("RESHAPE_1", plot_buf_1)
+# tf.summary.image("RESHAPE_2", plot_buf_2)
+# tf.summary.image("RESHAPE_3", plot_buf_3)
+# tf.summary.image("RESHAPE_4", plot_buf_4)
+# tf.summary.image("RESHAPE_5", plot_buf_5)
+# tf.summary.image("RESHAPE_6", plot_buf_6)
+# tf.summary.image("RESHAPE_7", plot_buf_7)
+# tf.summary.image("RESHAPE_8", plot_buf_8)
+
+
+ summary_8 = tf.summary.image("RESHAPE_8", plot_buf_8)
+ summary_7 = tf.summary.image("RESHAPE_7", plot_buf_7)
+ summary_6 = tf.summary.image("RESHAPE_6", plot_buf_6)
+ summary_5 = tf.summary.image("RESHAPE_5", plot_buf_5)
+ summary_4 = tf.summary.image("RESHAPE_4", plot_buf_4)
+ summary_3 = tf.summary.image("RESHAPE_3", plot_buf_3)
+ summary_2 = tf.summary.image("RESHAPE_2", plot_buf_2)
+ summary_1 = tf.summary.image("RESHAPE_1", plot_buf_1)
+
+
+# summary_op = tf.summary.image("RESHAPE_4", plot_buf_4)
+# with tf.Session() as sess:
+# y_pred,1_summary,2_summary = sess.run([Y_pred,summary_op_1,summary_op_2], feed_dict={X: x_sample})
+
+ y_pred,summary_pred_1,summary_pred_2,summary_pred_3,summary_pred_4,summary_pred_5,summary_pred_6,summary_pred_7,summary_pred_8 = sess.run([Y_pred,summary_1,summary_2,summary_3,summary_4,summary_5,summary_6,summary_7,summary_8], feed_dict={X: x_sample})
+
+# Write summary tf.summary.FileWriter
+ writer = tf.summary.FileWriter('./logs')
+
+
+ writer.add_summary(summary_pred_1)
+ writer.add_summary(summary_pred_2)
+ writer.add_summary(summary_pred_3)
+ writer.add_summary(summary_pred_4)
+ writer.add_summary(summary_pred_5)
+ writer.add_summary(summary_pred_6)
+ writer.add_summary(summary_pred_7)
+ writer.add_summary(summary_pred_8)
+
+ writer.close()
+
+# sys.exit(). sys.exit()
+ ###### to visualize
+ th = 0.25
+ y_pred[y_pred>=th]=1
+ y_pred[y_pred
Date: Wed, 6 Apr 2022 08:48:18 +0000
Subject: [PATCH 10/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/LICENSE?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE | 308 ------------------
1 file changed, 308 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE
deleted file mode 100644
index 6478627e0..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE
+++ /dev/null
@@ -1,308 +0,0 @@
-MIT License
-
-Copyright (c) 2020 Ajith Balakrishnan
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-
-
-#============================================================================
-Apache License
- Version 2.0, January 2004
- http://www.apache.org/licenses/
-
- TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
-
- 1. Definitions.
-
- "License" shall mean the terms and conditions for use, reproduction,
- and distribution as defined by Sections 1 through 9 of this document.
-
- "Licensor" shall mean the copyright owner or entity authorized by
- the copyright owner that is granting the License.
-
- "Legal Entity" shall mean the union of the acting entity and all
- other entities that control, are controlled by, or are under common
- control with that entity. For the purposes of this definition,
- "control" means (i) the power, direct or indirect, to cause the
- direction or management of such entity, whether by contract or
- otherwise, or (ii) ownership of fifty percent (50%) or more of the
- outstanding shares, or (iii) beneficial ownership of such entity.
-
- "You" (or "Your") shall mean an individual or Legal Entity
- exercising permissions granted by this License.
-
- "Source" form shall mean the preferred form for making modifications,
- including but not limited to software source code, documentation
- source, and configuration files.
-
- "Object" form shall mean any form resulting from mechanical
- transformation or translation of a Source form, including but
- not limited to compiled object code, generated documentation,
- and conversions to other media types.
-
- "Work" shall mean the work of authorship, whether in Source or
- Object form, made available under the License, as indicated by a
- copyright notice that is included in or attached to the work
- (an example is provided in the Appendix below).
-
- "Derivative Works" shall mean any work, whether in Source or Object
- form, that is based on (or derived from) the Work and for which the
- editorial revisions, annotations, elaborations, or other modifications
- represent, as a whole, an original work of authorship. For the purposes
- of this License, Derivative Works shall not include works that remain
- separable from, or merely link (or bind by name) to the interfaces of,
- the Work and Derivative Works thereof.
-
- "Contribution" shall mean any work of authorship, including
- the original version of the Work and any modifications or additions
- to that Work or Derivative Works thereof, that is intentionally
- submitted to Licensor for inclusion in the Work by the copyright owner
- or by an individual or Legal Entity authorized to submit on behalf of
- the copyright owner. For the purposes of this definition, "submitted"
- means any form of electronic, verbal, or written communication sent
- to the Licensor or its representatives, including but not limited to
- communication on electronic mailing lists, source code control systems,
- and issue tracking systems that are managed by, or on behalf of, the
- Licensor for the purpose of discussing and improving the Work, but
- excluding communication that is conspicuously marked or otherwise
- designated in writing by the copyright owner as "Not a Contribution."
-
- "Contributor" shall mean Licensor and any individual or Legal Entity
- on behalf of whom a Contribution has been received by Licensor and
- subsequently incorporated within the Work.
-
- 2. Grant of Copyright License. Subject to the terms and conditions of
- this License, each Contributor hereby grants to You a perpetual,
- worldwide, non-exclusive, no-charge, royalty-free, irrevocable
- copyright license to reproduce, prepare Derivative Works of,
- publicly display, publicly perform, sublicense, and distribute the
- Work and such Derivative Works in Source or Object form.
-
- 3. Grant of Patent License. Subject to the terms and conditions of
- this License, each Contributor hereby grants to You a perpetual,
- worldwide, non-exclusive, no-charge, royalty-free, irrevocable
- (except as stated in this section) patent license to make, have made,
- use, offer to sell, sell, import, and otherwise transfer the Work,
- where such license applies only to those patent claims licensable
- by such Contributor that are necessarily infringed by their
- Contribution(s) alone or by combination of their Contribution(s)
- with the Work to which such Contribution(s) was submitted. If You
- institute patent litigation against any entity (including a
- cross-claim or counterclaim in a lawsuit) alleging that the Work
- or a Contribution incorporated within the Work constitutes direct
- or contributory patent infringement, then any patent licenses
- granted to You under this License for that Work shall terminate
- as of the date such litigation is filed.
-
- 4. Redistribution. You may reproduce and distribute copies of the
- Work or Derivative Works thereof in any medium, with or without
- modifications, and in Source or Object form, provided that You
- meet the following conditions:
-
- (a) You must give any other recipients of the Work or
- Derivative Works a copy of this License; and
-
- (b) You must cause any modified files to carry prominent notices
- stating that You changed the files; and
-
- (c) You must retain, in the Source form of any Derivative Works
- that You distribute, all copyright, patent, trademark, and
- attribution notices from the Source form of the Work,
- excluding those notices that do not pertain to any part of
- the Derivative Works; and
-
- (d) If the Work includes a "NOTICE" text file as part of its
- distribution, then any Derivative Works that You distribute must
- include a readable copy of the attribution notices contained
- within such NOTICE file, excluding those notices that do not
- pertain to any part of the Derivative Works, in at least one
- of the following places: within a NOTICE text file distributed
- as part of the Derivative Works; within the Source form or
- documentation, if provided along with the Derivative Works; or,
- within a display generated by the Derivative Works, if and
- wherever such third-party notices normally appear. The contents
- of the NOTICE file are for informational purposes only and
- do not modify the License. You may add Your own attribution
- notices within Derivative Works that You distribute, alongside
- or as an addendum to the NOTICE text from the Work, provided
- that such additional attribution notices cannot be construed
- as modifying the License.
-
- You may add Your own copyright statement to Your modifications and
- may provide additional or different license terms and conditions
- for use, reproduction, or distribution of Your modifications, or
- for any such Derivative Works as a whole, provided Your use,
- reproduction, and distribution of the Work otherwise complies with
- the conditions stated in this License.
-
- 5. Submission of Contributions. Unless You explicitly state otherwise,
- any Contribution intentionally submitted for inclusion in the Work
- by You to the Licensor shall be under the terms and conditions of
- this License, without any additional terms or conditions.
- Notwithstanding the above, nothing herein shall supersede or modify
- the terms of any separate license agreement you may have executed
- with Licensor regarding such Contributions.
-
- 6. Trademarks. This License does not grant permission to use the trade
- names, trademarks, service marks, or product names of the Licensor,
- except as required for reasonable and customary use in describing the
- origin of the Work and reproducing the content of the NOTICE file.
-
- 7. Disclaimer of Warranty. Unless required by applicable law or
- agreed to in writing, Licensor provides the Work (and each
- Contributor provides its Contributions) on an "AS IS" BASIS,
- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
- implied, including, without limitation, any warranties or conditions
- of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
- PARTICULAR PURPOSE. You are solely responsible for determining the
- appropriateness of using or redistributing the Work and assume any
- risks associated with Your exercise of permissions under this License.
-
- 8. Limitation of Liability. In no event and under no legal theory,
- whether in tort (including negligence), contract, or otherwise,
- unless required by applicable law (such as deliberate and grossly
- negligent acts) or agreed to in writing, shall any Contributor be
- liable to You for damages, including any direct, indirect, special,
- incidental, or consequential damages of any character arising as a
- result of this License or out of the use or inability to use the
- Work (including but not limited to damages for loss of goodwill,
- work stoppage, computer failure or malfunction, or any and all
- other commercial damages or losses), even if such Contributor
- has been advised of the possibility of such damages.
-
- 9. Accepting Warranty or Additional Liability. While redistributing
- the Work or Derivative Works thereof, You may choose to offer,
- and charge a fee for, acceptance of support, warranty, indemnity,
- or other liability obligations and/or rights consistent with this
- License. However, in accepting such obligations, You may act only
- on Your own behalf and on Your sole responsibility, not on behalf
- of any other Contributor, and only if You agree to indemnify,
- defend, and hold each Contributor harmless for any liability
- incurred by, or claims asserted against, such Contributor by reason
- of your accepting any such warranty or additional liability.
-
- END OF TERMS AND CONDITIONS
-
- APPENDIX: How to apply the Apache License to your work.
-
- To apply the Apache License to your work, attach the following
- boilerplate notice, with the fields enclosed by brackets "[]"
- replaced with your own identifying information. (Don't include
- the brackets!) The text should be enclosed in the appropriate
- comment syntax for the file format. We also recommend that a
- file or class name and description of purpose be included on the
- same "printed page" as the copyright notice for easier
- identification within third-party archives.
-
- Copyright [yyyy] [name of copyright owner]
-
- Licensed under the Apache License, Version 2.0 (the "License");
- you may not use this file except in compliance with the License.
- You may obtain a copy of the License at
-
- http://www.apache.org/licenses/LICENSE-2.0
-
- Unless required by applicable law or agreed to in writing, software
- distributed under the License is distributed on an "AS IS" BASIS,
- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- See the License for the specific language governing permissions and
- limitations under the License.
-
-------------------
-Files: third_party/compute_library/...
-
-MIT License
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-
-------------------
-Files: ACKNOWLEDGEMENTS
-LICENSE
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-1. Redistributions of source code must retain the above copyright notice, this
- list of conditions and the following disclaimer.
-
-2. Redistributions in binary form must reproduce the above copyright notice,
- this list of conditions and the following disclaimer in the documentation
- and/or other materials provided with the distribution.
-
- THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
- ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
- WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
- DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
- ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
- (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
- LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
- ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-------------------
-Files: third_party/hexagon
-
-Copyright (c) 2016-2019, The Linux Foundation. All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted (subject to the limitations in the
-disclaimer below) provided that the following conditions are met:
-
- * Redistributions of source code must retain the above copyright
- notice, this list of conditions and the following disclaimer.
-
- * Redistributions in binary form must reproduce the above
- copyright notice, this list of conditions and the following
- disclaimer in the documentation and/or other materials provided
- with the distribution.
-
- * Neither the name of The Linux Foundation nor the names of its
- contributors may be used to endorse or promote products derived
- from this software without specific prior written permission.
-
-NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE
-GRANTED BY THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT
-HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED
-WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
-MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
-IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
-ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
-DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
-GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
-INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
-IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
-OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
-IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
\ No newline at end of file
--
Gitee
From fc2003bcd3cd290f0e2c3663bceb445d50713bd1 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:48:23 +0000
Subject: [PATCH 11/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/README.?=
=?UTF-8?q?md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../Pix2Vox_ID1284_for_TensorFlow/README.md | 84 -------------------
1 file changed, 84 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md
deleted file mode 100644
index 11c24c143..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md
+++ /dev/null
@@ -1,84 +0,0 @@
-# 3D_Object_Reconstruction
-
-#### 基本信息
-发布者（Publisher）：Huawei
-
-应用领域（Application Domain）： 3D Object Reconstruction
-
-版本（Version）：1.2
-
-修改时间（Modified） ：2021.11.06
-
-框架（Framework）：TensorFlow 1.15.0
-
-模型格式（Model Format）：ckpt
-
-精度（Precision）：Mixed
-
-处理器（Processor）：昇腾910
-
-应用级别（Categories）：Official
-
-描述（Description）：基于TensorFlow框架的物体三维重建网络训练代码
-
-
-
-#### 概述
-A Novel Hybrid Ensemble Approach For 3D Object Reconstruction from Multi-View Monocular RGB images for Robotic Simulations.
-
-Code comes from: https://github.com/Ajithbalakrishnan/3D-Object-Reconstruction-from-Multi-View-Monocular-RGB-images
-
-参考论文：https://arxiv.org/pdf/1901.11153.pdf
-
-参考实现：
-
-适配昇腾 AI 处理器的实现：https://toscode.gitee.com/ascend/modelzoo/pulls/5278
-
-#### 算法架构
-
-
-
-#### 默认配置
-
-数据集
-
-ShapeNet rendered images http://cvgl.stanford.edu/data2/ShapeNetRendering.tgz
-
-ShapeNet voxelized models http://cvgl.stanford.edu/data2/ShapeNetVox32.tg
-
-运行环境
-
-python 3.5
-
-tensorflow 1.13.0
-
-numpy 1.13.3
-
-scipy 0.19.0
-
-matplotlib
-
-skimage
-
-PyMCubes
-
-超参数设置
-
-batchsize:2
-
-Learning rate(LR): 0.001/0.0005/0.0001
-
-Train epoch: 50
-
-#### 训练模型
-
-The code runs on HUAWEI's Modelarts. The main program is ./Code/modelarts_entry.py
-
-
-#### 其他说明
-
-1. Unzip the data set to ./Data/ShapeNetRendering and ./Data/ShapeNetVox32
-
-2. The training log on Modelarts has been saved in the log folder.
-
-3. Due to the limitation of running time, the program only trained on the airplane data set(02691156).
--
Gitee
From d5b77145439f4e0d1414bb62c1f0a5ecd576cef4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:48:30 +0000
Subject: [PATCH 12/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/modelzo?=
=?UTF-8?q?o=5Flevel.txt?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt | 6 ------
1 file changed, 6 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt
deleted file mode 100644
index 7b4ecc749..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-GPUStatus:OK
-NPUMigrationStatus:OK
-FuncStatus:OK
-PrecisionStatus:POK
-AutoTune:NOK
-PerfStatus:NOK
\ No newline at end of file
--
Gitee
From f5a3e1b64b61780f76cf6465ea2f2704ce490066 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:48:37 +0000
Subject: [PATCH 13/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/require?=
=?UTF-8?q?ments.txt?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt | 11 -----------
1 file changed, 11 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt
deleted file mode 100644
index 75fd48ace..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt
+++ /dev/null
@@ -1,11 +0,0 @@
-tensorflow==1.13.0
-
-numpy==1.13.3
-
-scipy==0.19.0
-
-matplotlib
-
-skimage
-
-PyMCubes
\ No newline at end of file
--
Gitee
From 0b3d5875a339112be8224348827ed2e03dc89349 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:49:27 +0000
Subject: [PATCH 14/16] =?UTF-8?q?=E6=8F=90=E4=BA=A4=E8=AE=AD=E7=BB=83?=
=?UTF-8?q?=E7=9B=B8=E5=85=B3=E6=96=87=E4=BB=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE | 308 +++++++++++
.../Pix2Vox_ID1284_for_TensorFlow/README.md | 178 +++++++
.../modelarts_entry_acc.py | 64 +++
.../modelzoo_level.txt | 6 +
.../requirements.txt | 11 +
.../cv/Pix2Vox_ID1284_for_TensorFlow/tools.py | 482 ++++++++++++++++++
.../cv/Pix2Vox_ID1284_for_TensorFlow/voxel.py | 89 ++++
7 files changed, 1138 insertions(+)
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelarts_entry_acc.py
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/tools.py
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/voxel.py
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE
new file mode 100644
index 000000000..6478627e0
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/LICENSE
@@ -0,0 +1,308 @@
+MIT License
+
+Copyright (c) 2020 Ajith Balakrishnan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+
+#============================================================================
+Apache License
+ Version 2.0, January 2004
+ http://www.apache.org/licenses/
+
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+ 1. Definitions.
+
+ "License" shall mean the terms and conditions for use, reproduction,
+ and distribution as defined by Sections 1 through 9 of this document.
+
+ "Licensor" shall mean the copyright owner or entity authorized by
+ the copyright owner that is granting the License.
+
+ "Legal Entity" shall mean the union of the acting entity and all
+ other entities that control, are controlled by, or are under common
+ control with that entity. For the purposes of this definition,
+ "control" means (i) the power, direct or indirect, to cause the
+ direction or management of such entity, whether by contract or
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
+ outstanding shares, or (iii) beneficial ownership of such entity.
+
+ "You" (or "Your") shall mean an individual or Legal Entity
+ exercising permissions granted by this License.
+
+ "Source" form shall mean the preferred form for making modifications,
+ including but not limited to software source code, documentation
+ source, and configuration files.
+
+ "Object" form shall mean any form resulting from mechanical
+ transformation or translation of a Source form, including but
+ not limited to compiled object code, generated documentation,
+ and conversions to other media types.
+
+ "Work" shall mean the work of authorship, whether in Source or
+ Object form, made available under the License, as indicated by a
+ copyright notice that is included in or attached to the work
+ (an example is provided in the Appendix below).
+
+ "Derivative Works" shall mean any work, whether in Source or Object
+ form, that is based on (or derived from) the Work and for which the
+ editorial revisions, annotations, elaborations, or other modifications
+ represent, as a whole, an original work of authorship. For the purposes
+ of this License, Derivative Works shall not include works that remain
+ separable from, or merely link (or bind by name) to the interfaces of,
+ the Work and Derivative Works thereof.
+
+ "Contribution" shall mean any work of authorship, including
+ the original version of the Work and any modifications or additions
+ to that Work or Derivative Works thereof, that is intentionally
+ submitted to Licensor for inclusion in the Work by the copyright owner
+ or by an individual or Legal Entity authorized to submit on behalf of
+ the copyright owner. For the purposes of this definition, "submitted"
+ means any form of electronic, verbal, or written communication sent
+ to the Licensor or its representatives, including but not limited to
+ communication on electronic mailing lists, source code control systems,
+ and issue tracking systems that are managed by, or on behalf of, the
+ Licensor for the purpose of discussing and improving the Work, but
+ excluding communication that is conspicuously marked or otherwise
+ designated in writing by the copyright owner as "Not a Contribution."
+
+ "Contributor" shall mean Licensor and any individual or Legal Entity
+ on behalf of whom a Contribution has been received by Licensor and
+ subsequently incorporated within the Work.
+
+ 2. Grant of Copyright License. Subject to the terms and conditions of
+ this License, each Contributor hereby grants to You a perpetual,
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+ copyright license to reproduce, prepare Derivative Works of,
+ publicly display, publicly perform, sublicense, and distribute the
+ Work and such Derivative Works in Source or Object form.
+
+ 3. Grant of Patent License. Subject to the terms and conditions of
+ this License, each Contributor hereby grants to You a perpetual,
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+ (except as stated in this section) patent license to make, have made,
+ use, offer to sell, sell, import, and otherwise transfer the Work,
+ where such license applies only to those patent claims licensable
+ by such Contributor that are necessarily infringed by their
+ Contribution(s) alone or by combination of their Contribution(s)
+ with the Work to which such Contribution(s) was submitted. If You
+ institute patent litigation against any entity (including a
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
+ or a Contribution incorporated within the Work constitutes direct
+ or contributory patent infringement, then any patent licenses
+ granted to You under this License for that Work shall terminate
+ as of the date such litigation is filed.
+
+ 4. Redistribution. You may reproduce and distribute copies of the
+ Work or Derivative Works thereof in any medium, with or without
+ modifications, and in Source or Object form, provided that You
+ meet the following conditions:
+
+ (a) You must give any other recipients of the Work or
+ Derivative Works a copy of this License; and
+
+ (b) You must cause any modified files to carry prominent notices
+ stating that You changed the files; and
+
+ (c) You must retain, in the Source form of any Derivative Works
+ that You distribute, all copyright, patent, trademark, and
+ attribution notices from the Source form of the Work,
+ excluding those notices that do not pertain to any part of
+ the Derivative Works; and
+
+ (d) If the Work includes a "NOTICE" text file as part of its
+ distribution, then any Derivative Works that You distribute must
+ include a readable copy of the attribution notices contained
+ within such NOTICE file, excluding those notices that do not
+ pertain to any part of the Derivative Works, in at least one
+ of the following places: within a NOTICE text file distributed
+ as part of the Derivative Works; within the Source form or
+ documentation, if provided along with the Derivative Works; or,
+ within a display generated by the Derivative Works, if and
+ wherever such third-party notices normally appear. The contents
+ of the NOTICE file are for informational purposes only and
+ do not modify the License. You may add Your own attribution
+ notices within Derivative Works that You distribute, alongside
+ or as an addendum to the NOTICE text from the Work, provided
+ that such additional attribution notices cannot be construed
+ as modifying the License.
+
+ You may add Your own copyright statement to Your modifications and
+ may provide additional or different license terms and conditions
+ for use, reproduction, or distribution of Your modifications, or
+ for any such Derivative Works as a whole, provided Your use,
+ reproduction, and distribution of the Work otherwise complies with
+ the conditions stated in this License.
+
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
+ any Contribution intentionally submitted for inclusion in the Work
+ by You to the Licensor shall be under the terms and conditions of
+ this License, without any additional terms or conditions.
+ Notwithstanding the above, nothing herein shall supersede or modify
+ the terms of any separate license agreement you may have executed
+ with Licensor regarding such Contributions.
+
+ 6. Trademarks. This License does not grant permission to use the trade
+ names, trademarks, service marks, or product names of the Licensor,
+ except as required for reasonable and customary use in describing the
+ origin of the Work and reproducing the content of the NOTICE file.
+
+ 7. Disclaimer of Warranty. Unless required by applicable law or
+ agreed to in writing, Licensor provides the Work (and each
+ Contributor provides its Contributions) on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ implied, including, without limitation, any warranties or conditions
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+ PARTICULAR PURPOSE. You are solely responsible for determining the
+ appropriateness of using or redistributing the Work and assume any
+ risks associated with Your exercise of permissions under this License.
+
+ 8. Limitation of Liability. In no event and under no legal theory,
+ whether in tort (including negligence), contract, or otherwise,
+ unless required by applicable law (such as deliberate and grossly
+ negligent acts) or agreed to in writing, shall any Contributor be
+ liable to You for damages, including any direct, indirect, special,
+ incidental, or consequential damages of any character arising as a
+ result of this License or out of the use or inability to use the
+ Work (including but not limited to damages for loss of goodwill,
+ work stoppage, computer failure or malfunction, or any and all
+ other commercial damages or losses), even if such Contributor
+ has been advised of the possibility of such damages.
+
+ 9. Accepting Warranty or Additional Liability. While redistributing
+ the Work or Derivative Works thereof, You may choose to offer,
+ and charge a fee for, acceptance of support, warranty, indemnity,
+ or other liability obligations and/or rights consistent with this
+ License. However, in accepting such obligations, You may act only
+ on Your own behalf and on Your sole responsibility, not on behalf
+ of any other Contributor, and only if You agree to indemnify,
+ defend, and hold each Contributor harmless for any liability
+ incurred by, or claims asserted against, such Contributor by reason
+ of your accepting any such warranty or additional liability.
+
+ END OF TERMS AND CONDITIONS
+
+ APPENDIX: How to apply the Apache License to your work.
+
+ To apply the Apache License to your work, attach the following
+ boilerplate notice, with the fields enclosed by brackets "[]"
+ replaced with your own identifying information. (Don't include
+ the brackets!) The text should be enclosed in the appropriate
+ comment syntax for the file format. We also recommend that a
+ file or class name and description of purpose be included on the
+ same "printed page" as the copyright notice for easier
+ identification within third-party archives.
+
+ Copyright [yyyy] [name of copyright owner]
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+
+------------------
+Files: third_party/compute_library/...
+
+MIT License
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+------------------
+Files: ACKNOWLEDGEMENTS
+LICENSE
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+ list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+ this list of conditions and the following disclaimer in the documentation
+ and/or other materials provided with the distribution.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+------------------
+Files: third_party/hexagon
+
+Copyright (c) 2016-2019, The Linux Foundation. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted (subject to the limitations in the
+disclaimer below) provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+
+ * Redistributions in binary form must reproduce the above
+ copyright notice, this list of conditions and the following
+ disclaimer in the documentation and/or other materials provided
+ with the distribution.
+
+ * Neither the name of The Linux Foundation nor the names of its
+ contributors may be used to endorse or promote products derived
+ from this software without specific prior written permission.
+
+NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE
+GRANTED BY THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT
+HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED
+WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
+IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
+OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
+IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
\ No newline at end of file
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md
new file mode 100644
index 000000000..e6025a057
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/README.md
@@ -0,0 +1,178 @@
+- [基本信息](#基本信息.md)
+- [概述](#概述.md)
+- [训练环境准备](#训练环境准备.md)
+- [快速上手](#快速上手.md)
+- [训练结果](#训练结果.md)
+- [高级参考](#高级参考.md)
+基本信息
+
+**发布者（Publisher）：Huawei**
+
+**应用领域（Application Domain）：3D Object Reconstruction**
+
+**版本（Version）：1.2**
+
+**修改时间（Modified） ：2022.2.11**
+
+**大小（Size）：110KB**
+
+**框架（Framework）：TensorFlow 1.15.0**
+
+**模型格式（Model Format）：ckpt**
+
+**精度（Precision）：Mixed**
+
+**处理器（Processor）：昇腾910**
+
+**应用级别（Categories）：Official**
+
+**描述（Description）：基于TensorFlow框架的物体三维重建网络训练代码**
+
+概述
+
+ A Novel Hybrid Ensemble Approach For 3D Object Reconstruction from Multi-View Monocular RGB images for Robotic Simulations.
+
+- 参考论文：
+
+ https://arxiv.org/pdf/1901.11153.pdf
+
+- 参考实现：
+
+ https://github.com/Ajithbalakrishnan/3D-Object-Reconstruction-from-Multi-View-Monocular-RGB-images
+
+- 适配昇腾 AI 处理器的实现：
+
+
+ https://toscode.gitee.com/ascend/modelzoo/pulls/5278
+
+
+## 默认配置
+
+- 训练数据集预处理（以ShapeNet训练集为例，仅作为用户参考示例）：
+
+ - 图像的输入尺寸为137*137
+ - 图像输入格式：png
+
+- 测试数据集预处理（以ShapeNet验证集为例，仅作为用户参考示例）
+
+ - 图像的输入尺寸为137*137
+ - 图像输入格式：png
+
+- 数据集获取：
+
+ http://cvgl.stanford.edu/data2/ShapeNetRendering.tgz
+ http://cvgl.stanford.edu/data2/ShapeNetVox32.tgz
+
+- 训练超参
+
+ - Batch size： 2
+ - Train epoch: 50
+ - Train step: 80900
+
+
+## 支持特性
+
+| 特性列表 | 是否支持 |
+|-------|------|
+| 分布式训练 | 否 |
+| 混合精度 | 是 |
+| 并行数据 | 是 |
+
+## 混合精度训练
+
+昇腾910 AI处理器提供自动混合精度功能，可以针对全网中float32数据类型的算子，按照内置的优化策略，自动将部分float32的算子降低精度到float16，从而在精度损失很小的情况下提升系统性能并减少内存使用。
+
+## 开启混合精度
+
+脚本已默认开启混合精度。
+
+训练环境准备
+
+1. 硬件环境准备请参见各硬件产品文档"[驱动和固件安装升级指南]( https://support.huawei.com/enterprise/zh/category/ai-computing-platform-pid-1557196528909)"。需要在硬件设备上安装与CANN版本配套的固件与驱动。
+2. 代码使用的华为镜像为ascend-share/5.0.4.alpha005_tensorflow-ascend910-cp37-euleros2.8-aarch64-training:1.15.0-21.0.2_1224。
+
+
+
+快速上手
+
+- 数据集准备
+1. 模型训练使用ShapeNet数据集，数据集请自行获取。
+
+## 模型训练
+
+- 单击“立即下载”，并选择合适的下载方式下载源码包。
+
+- 启动训练之前，首先要配置程序运行相关环境变量。
+
+ 环境变量配置信息参见：
+
+ [Ascend 910训练平台环境变量设置](https://gitee.com/ascend/modelzoo/wikis/Ascend%20910%E8%AE%AD%E7%BB%83%E5%B9%B3%E5%8F%B0%E7%8E%AF%E5%A2%83%E5%8F%98%E9%87%8F%E8%AE%BE%E7%BD%AE?sort_id=3148819)
+
+- 单卡训练
+
+ 1. 配置训练参数。
+
+ 首先在脚本main_AttSets.py中，配置batch_size、epochs、total_mv等参数，请用户根据实际路径配置data_path，或者在启动训练的命令行中以参数形式下发。
+
+ ```
+ batch_size=2
+ total_mv=24
+ epochs=50
+ ```
+
+ 2. 启动训练。
+
+ 启动单卡训练 （脚本为Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh）
+
+ ```
+ bash train_full_1p.sh
+ ```
+
+训练结果
+
+- 精度结果比对
+
+|精度指标项|论文发布|GPU实测|NPU实测|
+|---|---|---|---|
+|ACC|NA|0.8184|0.8289|
+
+- 性能结果比对
+
+|性能指标项|论文发布|GPU实测|NPU实测|
+|---|---|---|---|
+|FPS|NA|34.12|14.15|
+
+
+高级参考
+
+## 脚本和示例代码
+
+```
+├── main_AttSets.py //网络训练代码
+├── demo_AttSets.py //网络测试代码
+├── binvox_rw.py //体素模型预处理代码
+├── export_obj.py //模型加载代码
+├── tools.py //自定义工具包
+├── voxel.py //体素模型预处理代码
+├── npu_train.sh //npu训练脚本
+├── README.md //代码说明文档
+├── requirements.tx //训练python依赖列表
+├── test
+│ ├──train_full_1p.sh //单卡全量训练启动脚本
+```
+
+## 脚本参数
+
+```
+--data_path 数据集路径
+--batch_size 每个NPU的batch size，默认：2
+--att_lr、ref_lr 学习率，默认：0.0002/0.0001/0.00005
+--epochs 训练epcoh数量，默认：50
+```
+
+## 训练过程
+
+1. 通过“模型训练”中的训练指令启动单卡卡训练。
+
+2. 参考脚本的模型存储路径为./Model/train_mod。
+
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelarts_entry_acc.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelarts_entry_acc.py
new file mode 100644
index 000000000..ae7b7985a
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelarts_entry_acc.py
@@ -0,0 +1,64 @@
+#
+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+import os
+import argparse
+import sys
+
+# 解析输入参数data_url
+parser = argparse.ArgumentParser()
+parser.add_argument("--data_url", type=str, default="/home/ma-user/modelarts/inputs/data_url_0")
+parser.add_argument("--train_url", type=str, default="/home/ma-user/modelarts/outputs/train_url_0/")
+config = parser.parse_args()
+
+print("[CANN-Modelzoo] code_dir path is [%s]" % (sys.path[0]))
+code_dir = sys.path[0]
+os.chdir(code_dir)
+print("[CANN-Modelzoo] work_dir path is [%s]" % (os.getcwd()))
+
+print("[CANN-Modelzoo] before train - list my run files:")
+os.system("ls -al /usr/local/Ascend/ascend-toolkit/")
+
+print("[CANN-Modelzoo] before train - list my dataset files:")
+os.system("ls -al %s" % config.data_url)
+
+print("[CANN-Modelzoo] start run train shell")
+# 设置sh文件格式为linux可执行
+os.system("dos2unix ./test/*")
+
+# 执行train_full_1p.sh或者train_performance_1p.sh，需要用户自己指定
+# full和performance的差异，performance只需要执行很少的step，控制在15分钟以内，主要关注性能FPS
+os.system("bash ./test/train_full_1p.sh --data_path=%s --output_path=%s " % (config.data_url, config.train_url))
+
+print("[CANN-Modelzoo] finish run train shell")
+
+# 将当前执行目录所有文件拷贝到obs的output进行备份
+print("[CANN-Modelzoo] after train - list my output files:")
+os.system("cp -r %s %s " % (code_dir, config.train_url))
+os.system("ls -al %s" % config.train_url)
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt
new file mode 100644
index 000000000..cef7e6440
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/modelzoo_level.txt
@@ -0,0 +1,6 @@
+GPUStatus:OK
+NPUMigrationStatus:OK
+FuncStatus:OK
+PrecisionStatus:OK
+AutoTune:NOK
+PerfStatus:NOK
\ No newline at end of file
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt
new file mode 100644
index 000000000..8b716f006
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/requirements.txt
@@ -0,0 +1,11 @@
+tensorflow==1.13.0
+
+numpy==1.13.3
+
+scipy==0.19.0
+
+matplotlib
+
+skimage
+
+PyMCubes
\ No newline at end of file
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/tools.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/tools.py
new file mode 100644
index 000000000..6054ceb1e
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/tools.py
@@ -0,0 +1,482 @@
+#
+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+from npu_bridge.npu_init import *
+import tensorflow as tf
+from PIL import Image
+import numpy as np
+import os
+import sys
+import random
+import matplotlib.pyplot as plt
+import matplotlib.gridspec as gridspec
+from mpl_toolkits import mplot3d
+import re
+import binvox_rw as binvox_rw
+#################################################################
+from voxel import voxel2obj
+#from PyMCubes import mcubes
+from export_obj import export_obj
+#################################################################
+class Data:
+ def __init__(self, config):
+ self.config = config
+ self.batch_size = config['batch_size']
+ self.train_batch_index = 0
+ self.test_batch_index_sq = 0
+ self.cat_names = config['cat_names']
+ self.total_mv = config['total_mv']
+ self.cat_test_1st_index = None
+
+ self.X_rgb_train_files_ori, self.Y_vox_train_files_ori, self.X_rgb_test_files_ori, self.Y_vox_test_files_ori \
+ =self.load_X_Y_files_paths_all(self.cat_names)
+
+ print ('X_rgb_train_files_ori:', len(self.X_rgb_train_files_ori))
+ print ('X_rgb_test_files_ori:', len(self.X_rgb_test_files_ori))
+
+ @staticmethod
+ def crop_center(im, new_height, new_width):
+# print('crop_center')
+ height = im.shape[0] # Get dimensions
+ width = im.shape[1]
+ left = (width - new_width) // 2
+ top = (height - new_height) // 2
+ right = (width + new_width) // 2
+ bottom = (height + new_height) // 2
+ return im[top:bottom, left:right]
+
+ @staticmethod
+ def add_random_color_background(im, color_range):
+# print('add_random_color_background')
+ r, g, b = [np.random.randint(color_range[i][0], color_range[i][1] + 1) for i in range(3)]
+# print (r)
+# print (g)
+# print (b)
+ if isinstance(im, Image.Image):
+ im = np.array(im)
+# print (im.shape)
+# b = np.array([3]) ###
+# np.concatenate((im, '3')) ##
+# im=np.append(im,3) ###
+# print (im.shape) ###
+ if im.shape[2] > 3:
+
+ # If the image has the alpha channel, add the background
+ alpha = (np.expand_dims(im[:, :, 3], axis=2) == 0).astype(np.float)
+
+ im = im[:, :, :3]
+ bg_color = np.array([[[r, g, b]]])
+ im = alpha * bg_color + (1 - alpha) * im
+ return im
+
+ @staticmethod
+ def image_transform(img, crop_x, crop_y, crop_loc=None, color_tint=None):
+# print('image_transform')
+ RANDOM_CROP = True
+ # Slight translation
+ if RANDOM_CROP and not crop_loc:
+ crop_loc = [np.random.randint(0, crop_y), np.random.randint(0, crop_x)]
+
+ if crop_loc:
+ cr, cc = crop_loc
+ height, width, _ = img.shape
+ img_h = height - crop_y
+ img_w = width - crop_x
+ img = img[cr:cr + img_h, cc:cc + img_w]
+ # depth = depth[cr:cr+img_h, cc:cc+img_w]
+
+ FLIP = True
+ if FLIP and np.random.rand() > 0.5:
+ img = img[:, ::-1, ...]
+ return img
+
+ @staticmethod
+ def preprocess_img(im, train):
+# print('processing_img')
+ # add random background
+ TRAIN_NO_BG_COLOR_RANGE = [[225, 255], [225, 255], [225, 255]]
+ TEST_NO_BG_COLOR_RANGE = [[240, 240], [240, 240], [240, 240]]
+ im = Data.add_random_color_background(im, TRAIN_NO_BG_COLOR_RANGE if train else TEST_NO_BG_COLOR_RANGE)
+
+ # If the image has alpha channel, remove it.
+ CONST_IMG_W = 127
+ CONST_IMG_H = 127
+ im_rgb = np.array(im)[:, :, :3].astype(np.float32)
+ if train:
+ # Data augmentation
+ PAD_X = 10
+ PAD_Y = 10
+ t_im = Data.image_transform(im_rgb, PAD_X, PAD_Y)
+ else:
+ t_im = Data.crop_center(im_rgb, CONST_IMG_H, CONST_IMG_W)
+
+ # Scale image
+ t_im = t_im / 255.
+ return t_im
+
+ @staticmethod
+ def load_single_X_rgb_r2n2(img_path, train):
+# print('load_single_X_rgb_r2n2')
+ im = Image.open(img_path)
+
+ t_im = Data.preprocess_img(im, train=train)
+ # plt.figure()
+ # plt.imshow(t_im)
+ return t_im
+
+ @staticmethod
+ def load_single_Y_vox(vox_path):
+# print('load_single_Y_vox')
+ with open(vox_path, 'rb') as ff:
+ vox = binvox_rw.read_as_3d_array(ff)
+ vox_grid = vox.data.astype(int)
+
+ #Data.plotFromVoxels(vox_grid)
+ return vox_grid
+
+ @staticmethod
+ def load_X_Y_files_paths(X_cat_folder, Y_cat_folder):
+# print('load_X_Y_files_paths')
+ X_obj_folders=[X_f for X_f in sorted(os.listdir(X_cat_folder))]
+ Y_obj_folders=[Y_f for Y_f in sorted(os.listdir(Y_cat_folder))]
+ if len(X_obj_folders) != len(Y_obj_folders):
+ print ('Files are inconsistent in:', X_cat_folder, 'and', Y_cat_folder)
+
+ #### split train/test, according to 3D-R2N2 paper
+ train_num = int(0.8*len(X_obj_folders))
+ idx = list(range(train_num))
+
+ test_objs = []
+ X_train_obj_folders =[]; Y_train_obj_folders=[]
+ X_test_obj_folders =[]; Y_test_obj_folders=[]
+ for i in range(len(X_obj_folders)):
+ obj_na = X_obj_folders[i]
+ if obj_na not in Y_obj_folders:
+ print ('inconsistent single obj ignored')
+ continue
+ if i in idx:
+ X_train_obj_folders.append(X_cat_folder+obj_na+'/')
+ Y_train_obj_folders.append(Y_cat_folder+obj_na+'/')
+ else:
+ X_test_obj_folders.append(X_cat_folder+obj_na+'/')
+ Y_test_obj_folders.append(Y_cat_folder+obj_na+'/')
+ test_objs.append(obj_na)
+ print ('train objs:', len(X_train_obj_folders))
+ print ('test objs:', len(X_test_obj_folders))
+
+ #########
+ def load_x_y_files(X_obj_fo, Y_obj_fo):
+# print('load_x_y_files')
+ X_files_paths = []
+ Y_files_paths = []
+ for j in range(len(X_obj_fo)):
+ if X_obj_fo[j][-5:] not in Y_obj_fo[j]:
+ print ('inconsistent single obj exit')
+ exit()
+ for xf in sorted(os.listdir(X_obj_fo[j]+'rendering/')):
+ if '.png' in xf:
+ X_files_paths.append(X_obj_fo[j]+'rendering/'+xf)
+ Y_files_paths.append(Y_obj_fo[j]+'model.binvox')
+ return X_files_paths, Y_files_paths
+ #########
+
+ X_train_files, Y_train_files = load_x_y_files(X_train_obj_folders, Y_train_obj_folders)
+ X_test_files, Y_test_files = load_x_y_files(X_test_obj_folders, Y_test_obj_folders)
+
+ return X_train_files, Y_train_files,X_test_files,Y_test_files
+
+ @staticmethod
+ def load_X_Y_rgb_vox(X_files_full_path, Y_files_full_path, train):
+# print('load_X_Y_rgb_vox')
+ if len(X_files_full_path) != len(Y_files_full_path):
+ print ('load_X_Y_rgb_vox error!')
+ exit()
+ X_rgb = []
+ Y_vox = []
+ for X_f, Y_f in zip(X_files_full_path, Y_files_full_path):
+ na = re.split('/', X_f)[-3]
+ if na not in Y_f:
+ print ('X Y rgb vox file not consistent!')
+ exit()
+ rgb = Data.load_single_X_rgb_r2n2(X_f, train=train)
+ X_rgb.append(rgb)
+
+ vox = Data.load_single_Y_vox(Y_f)
+ Y_vox.append(vox)
+
+ X_rgb = np.asarray(X_rgb)
+ Y_vox = np.asarray(Y_vox)
+ return X_rgb, Y_vox
+
+ def test_listdir(self, path):
+ print('当前执行路径', os.getcwd())
+ for file in os.listdir(path):
+ file_path = os.path.join(path, file)
+ print(file_path)
+
+ def load_X_Y_files_paths_all(self,cat_names):
+# print('load_X_Y_files_paths_all')
+ x_rgb_str='X_rgb_'
+ y_vox_str='Y_vox_'
+
+ X_train_files_paths_all=[]
+ Y_trian_files_paths_all=[]
+ X_test_files_paths_all=[]
+ Y_test_files_paths_all=[]
+ self.cat_test_1st_index = [0]
+ for name in cat_names:
+ self.test_listdir(os.getcwd())
+ print ('loading files:', name)
+ X_rgb_folder = self.config[x_rgb_str+name]
+ Y_vox_folder = self.config[y_vox_str+name]
+
+ X_train_files, Y_train_files, X_test_files, Y_test_files = self.load_X_Y_files_paths(X_rgb_folder, Y_vox_folder)
+ self.cat_test_1st_index.append(len(X_test_files))
+
+ for X_rgb_f, Y_vox_f in zip(X_train_files, Y_train_files):
+ X_train_files_paths_all.append(X_rgb_f)
+ Y_trian_files_paths_all.append(Y_vox_f)
+ for X_rgb_f, Y_vox_f in zip(X_test_files, Y_test_files):
+ X_test_files_paths_all.append(X_rgb_f)
+ Y_test_files_paths_all.append(Y_vox_f)
+
+ return X_train_files_paths_all, Y_trian_files_paths_all,X_test_files_paths_all,Y_test_files_paths_all
+
+ ################################
+ def load_X_Y_train_next_batch(self, train_mv):
+# print('load_X_Y_train_next_batch')
+ X_rgb_files = self.X_rgb_train_files[self.batch_size*self.train_batch_index*train_mv:self.batch_size*(self.train_batch_index+1)*train_mv]
+ Y_vox_files = self.Y_vox_train_files[self.batch_size*self.train_batch_index*train_mv:self.batch_size*(self.train_batch_index+1)*train_mv]
+ self.train_batch_index +=1
+
+ X_rgb, Y_vox = self.load_X_Y_rgb_vox(X_rgb_files, Y_vox_files, train=True)
+ X = []
+ Y = []
+ for b in range(self.batch_size):
+ X.append(X_rgb[b * train_mv:(b + 1) * train_mv, :])
+ Y.append(Y_vox[b * train_mv, :, :, :])
+ X = np.asarray(X)
+ Y = np.asarray(Y)
+ return X, Y
+
+ def load_X_Y_test_next_batch(self, test_mv):
+# print('load_X_Y_test_next_batch')
+ num = self.total_mv
+ idx = random.sample(range(len(self.X_rgb_test_files_ori)//num), self.batch_size) ##############added extra '/'
+ X_rgb_files = []
+ Y_vox_files =[]
+
+ for i in idx:
+ tp1 = self.X_rgb_test_files_ori[i*num:(i+1)*num]
+ tp2 = self.Y_vox_test_files_ori[i*num:(i+1)*num]
+ for x, y in zip(tp1, tp2):
+ X_rgb_files.append(x)
+ Y_vox_files.append(y)
+ X_rgb_batch, Y_vox_batch = self.load_X_Y_rgb_vox(X_rgb_files, Y_vox_files, train=False)
+
+ X =[]
+ Y =[]
+ for b in range(self.batch_size):
+ tp1 = X_rgb_batch[b*num:(b+1)*num,:,:,:]
+ tp2 = Y_vox_batch[b*num:(b+1)*num,:,:,:]
+ idx2 = random.sample(range(num), test_mv)
+ X.append(tp1[idx2])
+ Y.append(tp2[idx2[0]])
+ X = np.asarray(X)
+ Y = np.asarray(Y)
+ return X, Y
+
+ def shuffle_train_files(self, ep, train_mv):
+# print('shuffle_train_files')
+ num = self.total_mv
+ X_rgb_new=[]; Y_vox_new=[]
+ self.train_batch_index = 0
+ X_rgb = self.X_rgb_train_files_ori
+ Y_vox = self.Y_vox_train_files_ori
+ index = list(range(int(len(X_rgb)/num)))
+ random.Random(ep).shuffle(index)
+
+ for i in index:
+ tp1 = X_rgb[i*num:(i+1)*num]
+ tp2 = Y_vox[i*num:(i+1)*num]
+
+ view_ind = list(range(num))
+ random.Random(i+100).shuffle(view_ind)
+ valid_view_num = int(num / train_mv) * train_mv
+ view_ind = view_ind[0:valid_view_num]
+
+ for j in view_ind:
+ x = tp1[j]
+ y = tp2[j]
+ na = re.split('/', x)[-3]
+ if na not in y:
+ print ('X Y rgb vox file not consistent!')
+ exit()
+ X_rgb_new.append(x)
+ Y_vox_new.append(y)
+ self.X_rgb_train_files = X_rgb_new
+ self.Y_vox_train_files = Y_vox_new
+ self.total_train_batch_num = int(len(self.X_rgb_train_files)/(self.batch_size*train_mv))
+
+ def shuffle_test_files(self, test_mv, seed):
+# print('shuffle_test_files')
+ X_rgb_new=[]; Y_vox_new=[]
+ X_rgb = self.X_rgb_test_files_ori
+ Y_vox = self.Y_vox_test_files_ori
+
+ num = self.total_mv
+ ###
+ cat_test_1st_index_new =[]
+ for k in self.cat_test_1st_index:
+ cat_test_1st_index_new.append( int(k/num)* int(num/test_mv)*test_mv)
+ self.cat_test_1st_index=cat_test_1st_index_new
+ ###
+ total_obj = int(len(X_rgb)/num)
+ for i in range(total_obj):
+ view_ind = list(range(num))
+ random.Random(i+seed).shuffle(view_ind)
+
+ valid_view_num = int(num/test_mv)*test_mv
+ view_ind = view_ind[0:valid_view_num]
+ for id in view_ind:
+ X_rgb_new.append(X_rgb[i*num+id])
+ Y_vox_new.append(Y_vox[i*num+id])
+
+ self.X_rgb_test_files = X_rgb_new
+ self.Y_vox_test_files = Y_vox_new
+
+class Ops:
+
+ @staticmethod
+ def lrelu(x, leak=0.2):
+ return tf.nn.leaky_relu(x,alpha=0.2,name=None)
+
+ @staticmethod
+ def relu(x):
+ return tf.nn.relu(x)
+
+ @staticmethod
+ def xxlu(x, label='relu'):
+ if label == 'relu':
+ return Ops.relu(x)
+ if label == 'lrelu':
+ return Ops.lrelu(x, leak=0.2)
+
+ @staticmethod
+ def variable_sum(var, name):
+ with tf.name_scope(name):
+ mean = tf.reduce_mean(var)
+ tf.summary.scalar('mean', mean)
+ stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
+ tf.summary.scalar('stddev', stddev)
+ tf.summary.scalar('max', tf.reduce_max(var))
+ tf.summary.scalar('min', tf.reduce_min(var))
+ tf.summary.histogram('histogram', var)
+
+ @staticmethod
+ def variable_count():
+ total_para = 0
+ for variable in tf.trainable_variables():
+ shape = variable.get_shape()
+ variable_para = 1
+ for dim in shape:
+ variable_para *= dim.value
+ total_para += variable_para
+ return total_para
+
+ @staticmethod
+ def fc(x, out_d, name):
+ xavier_init = tf.contrib.layers.xavier_initializer()
+ zero_init = tf.zeros_initializer()
+ in_d = x.get_shape()[1]
+ w = tf.get_variable(name + '_w', [in_d, out_d], initializer=xavier_init,dtype=tf.float32)
+ b = tf.get_variable(name + '_b', [out_d], initializer=zero_init,dtype=tf.float32)
+ y = tf.nn.bias_add(tf.matmul(x, w), b)
+ Ops.variable_sum(w, name)
+ return y
+
+ @staticmethod
+ def conv2d(x, k, out_c, str, name, pad='SAME'):
+ xavier_init = tf.contrib.layers.xavier_initializer()
+ zero_init = tf.zeros_initializer()
+ in_c = x.get_shape()[3]
+ w = tf.get_variable(name + '_w', [k, k, in_c, out_c], initializer=xavier_init, dtype=tf.float32)
+ b = tf.get_variable(name + '_b', [out_c], initializer=zero_init,dtype=tf.float32)
+
+ stride = [1, str, str, 1]
+ y = tf.nn.bias_add(tf.nn.conv2d(x, w, stride, pad), b)
+ Ops.variable_sum(w, name)
+ return y
+
+ @staticmethod
+ def maxpool2d(x, k, s,name, pad='SAME'):
+ ker = [1,k,k,1]
+ str = [1,s,s,1]
+ y = tf.nn.max_pool(x,ksize=ker, strides=str, padding=pad, name=name)
+ return y
+
+ @staticmethod
+ def conv3d(x, k, out_c, str, name, pad='SAME'):
+ xavier_init = tf.contrib.layers.xavier_initializer()
+ zero_init = tf.zeros_initializer()
+ in_c = x.get_shape()[4]
+ w = tf.get_variable(name + '_w', [k, k, k, in_c, out_c], initializer=xavier_init, dtype=tf.float32)
+ b = tf.get_variable(name + '_b', [out_c], initializer=zero_init, dtype=tf.float32)
+ stride = [1, str, str, str, 1]
+ y = tf.nn.bias_add(tf.nn.conv3d(x, w, stride, pad), b)
+ Ops.variable_sum(w, name)
+ return y
+
+ @staticmethod
+ def deconv3d(x, k, out_c, str, name, pad='SAME'):
+ xavier_init = tf.contrib.layers.xavier_initializer()
+ zero_init = tf.zeros_initializer()
+ bat = tf.shape(x)[0]
+ [_, in_d1, in_d2, in_d3, in_c] = x.get_shape()
+ in_d1 = int(in_d1); in_d2 = int(in_d2); in_d3 = int(in_d3); in_c = int(in_c)
+ w = tf.get_variable(name + '_w', [k, k, k, out_c, in_c], initializer=xavier_init, dtype=tf.float32)
+ b = tf.get_variable(name + '_b', [out_c], initializer=zero_init, dtype=tf.float32)
+
+ out_shape = [bat, in_d1 * str, in_d2 * str, in_d3 * str, out_c]
+ stride = [1, str, str, str, 1]
+ y = tf.nn.conv3d_transpose(x, w, output_shape=out_shape, strides=stride, padding=pad)
+ y = tf.nn.bias_add(y, b)
+ Ops.variable_sum(w, name)
+ return y
+
+ @staticmethod
+ def maxpool3d(x,k,s,name,pad='SAME'):
+ ker = [1,k,k,k,1]
+ str = [1,s,s,s,1]
+ y = tf.nn.max_pool3d(x,ksize=ker,strides=str,padding=pad, name=name)
+ return y
+
+
+
+
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/voxel.py b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/voxel.py
new file mode 100644
index 000000000..8bce75ecf
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/voxel.py
@@ -0,0 +1,89 @@
+#
+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+from npu_bridge.npu_init import *
+import numpy as np
+
+
+def evaluate_voxel_prediction(preds, gt, thresh):
+ preds_occupy = preds[:, 1, :, :] >= thresh
+ diff = np.sum(np.logical_xor(preds_occupy, gt[:, 1, :, :]))
+ intersection = np.sum(np.logical_and(preds_occupy, gt[:, 1, :, :]))
+ union = np.sum(np.logical_or(preds_occupy, gt[:, 1, :, :]))
+ num_fp = np.sum(np.logical_and(preds_occupy, gt[:, 0, :, :])) # false positive
+ num_fn = np.sum(np.logical_and(np.logical_not(preds_occupy), gt[:, 1, :, :])) # false negative
+ return np.array([diff, intersection, union, num_fp, num_fn])
+
+
+def voxel2mesh(voxels, surface_view):
+ cube_verts = [[0, 0, 0], [0, 0, 1], [0, 1, 0], [0, 1, 1], [1, 0, 0], [1, 0, 1], [1, 1, 0],
+ [1, 1, 1]] # 8 points
+
+ cube_faces = [[0, 1, 2], [1, 3, 2], [2, 3, 6], [3, 7, 6], [0, 2, 6], [0, 6, 4], [0, 5, 1],
+ [0, 4, 5], [6, 7, 5], [6, 5, 4], [1, 7, 3], [1, 5, 7]] # 12 face
+
+ cube_verts = np.array(cube_verts)
+ cube_faces = np.array(cube_faces) + 1
+
+ scale = 0.01
+ cube_dist_scale = 1.1
+ verts = []
+ faces = []
+ curr_vert = 0
+
+ positions = np.where(voxels > 0.3)
+ voxels[positions] = 1
+ for i, j, k in zip(*positions):
+ # identifies if current voxel has an exposed face
+ if not surface_view or np.sum(voxels[i-1:i+2, j-1:j+2, k-1:k+2]) < 27:
+ verts.extend(scale * (cube_verts + cube_dist_scale * np.array([[i, j, k]])))
+ faces.extend(cube_faces + curr_vert)
+ curr_vert += len(cube_verts)
+
+ return np.array(verts), np.array(faces)
+
+
+def write_obj(filename, verts, faces):
+ """ write the verts and faces on file."""
+ with open(filename, 'w') as f:
+ # write vertices
+ f.write('g\n# %d vertex\n' % len(verts))
+ for vert in verts:
+ f.write('v %f %f %f\n' % tuple(vert))
+
+ # write faces
+ f.write('# %d faces\n' % len(faces))
+ for face in faces:
+ f.write('f %d %d %d\n' % tuple(face))
+
+
+def voxel2obj(filename, pred, surface_view = True):
+ verts, faces = voxel2mesh(pred, surface_view)
+ write_obj(filename, verts, faces)
+
--
Gitee
From 840cda80023b6de94d4af1b6c3d8abae59d511db Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:50:01 +0000
Subject: [PATCH 15/16] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20Te?=
=?UTF-8?q?nsorFlow/contrib/cv/Pix2Vox=5FID1284=5Ffor=5FTensorFlow/test/tr?=
=?UTF-8?q?ain=5Ffull=5F1p.sh?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../test/train_full_1p.sh | 27 -------------------
1 file changed, 27 deletions(-)
delete mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
deleted file mode 100644
index 8ae2d297a..000000000
--- a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
+++ /dev/null
@@ -1,27 +0,0 @@
-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
\ No newline at end of file
--
Gitee
From 1195397db6c8892ee831fa4a40cdfe0f8a879d62 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=A2=9C=E4=BA=9A=E6=96=87?= <18013846960@189.cn>
Date: Wed, 6 Apr 2022 08:50:12 +0000
Subject: [PATCH 16/16] =?UTF-8?q?=E6=8F=90=E4=BA=A4=E8=AE=AD=E7=BB=83?=
=?UTF-8?q?=E7=9B=B8=E5=85=B3=E6=96=87=E4=BB=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
.../test/train_full_1p.sh | 181 ++++++++++++++++++
1 file changed, 181 insertions(+)
create mode 100644 TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
diff --git a/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
new file mode 100644
index 000000000..e092e968b
--- /dev/null
+++ b/TensorFlow/contrib/cv/Pix2Vox_ID1284_for_TensorFlow/test/train_full_1p.sh
@@ -0,0 +1,181 @@
+#!/bin/bash
+
+##########################################################
+#########第3行 至 100行，请一定不要、不要、不要修改##########
+#########第3行 至 100行，请一定不要、不要、不要修改##########
+#########第3行 至 100行，请一定不要、不要、不要修改##########
+##########################################################
+# shell脚本所在路径
+cur_path=`echo $(cd $(dirname $0);pwd)`
+
+# 判断当前shell是否是performance
+perf_flag=`echo $0 | grep performance | wc -l`
+
+# 当前执行网络的名称
+Network=`echo $(cd $(dirname $0);pwd) | awk -F"/" '{print $(NF-1)}'`
+
+export RANK_SIZE=1
+export RANK_ID=0
+export JOB_ID=10087
+
+# 路径参数初始化
+data_path=""
+output_path=""
+
+# 帮助信息，不需要修改
+if [[ $1 == --help || $1 == -h ]];then
+ echo"usage:./train_performance_1P.sh "
+ echo " "
+ echo "parameter explain:
+ --data_path # dataset of training
+ --output_path # output of training
+ --train_steps # max_step for training
+ --train_epochs # max_epoch for training
+ --batch_size # batch size
+ -h/--help show help message
+ "
+ exit 1
+fi
+
+# 参数校验，不需要修改
+for para in $*
+do
+ if [[ $para == --data_path* ]];then
+ data_path=`echo ${para#*=}`
+ elif [[ $para == --output_path* ]];then
+ output_path=`echo ${para#*=}`
+ elif [[ $para == --train_steps* ]];then
+ train_steps=`echo ${para#*=}`
+ elif [[ $para == --train_epochs* ]];then
+ train_epochs=`echo ${para#*=}`
+ elif [[ $para == --batch_size* ]];then
+ batch_size=`echo ${para#*=}`
+ fi
+done
+
+# 校验是否传入data_path,不需要修改
+if [[ $data_path == "" ]];then
+ echo "[Error] para \"data_path\" must be config"
+ exit 1
+fi
+
+# 校验是否传入output_path,不需要修改
+if [[ $output_path == "" ]];then
+ output_path="./test/output/${ASCEND_DEVICE_ID}"
+fi
+
+# 设置打屏日志文件名，请保留，文件名为${print_log}
+print_log="./test/output/${ASCEND_DEVICE_ID}/train_${ASCEND_DEVICE_ID}.log"
+etp_flag=${etp_running_flag}
+if [ x"${etp_flag}" != xtrue ];
+then
+ echo "running without etp..."
+ print_log_name=`ls /home/ma-user/modelarts/log/ | grep proc-rank`
+ print_log="/home/ma-user/modelarts/log/${print_log_name}"
+fi
+echo ${print_log}
+
+CaseName=""
+function get_casename()
+{
+ if [ x"${perf_flag}" = x1 ];
+ then
+ CaseName=${Network}_bs${batch_size}_${RANK_SIZE}'p'_'perf'
+ else
+ CaseName=${Network}_bs${batch_size}_${RANK_SIZE}'p'_'acc'
+ fi
+}
+
+# 跳转到code目录
+cd ${cur_path}/../
+rm -rf ./test/output/${ASCEND_DEVICE_ID}
+mkdir -p ./test/output/${ASCEND_DEVICE_ID}
+
+# 训练开始时间记录，不需要修改
+start_time=$(date +%s)
+##########################################################
+#########第3行 至 100行，请一定不要、不要、不要修改##########
+#########第3行 至 100行，请一定不要、不要、不要修改##########
+#########第3行 至 100行，请一定不要、不要、不要修改##########
+##########################################################
+
+#=========================================================
+#=========================================================
+#========训练执行命令，需要根据您的网络进行修改==============
+#=========================================================
+#=========================================================
+# 您的训练数据集在${data_path}路径下，请直接使用这个变量获取
+# 您的训练输出目录在${output_path}路径下，请直接使用这个变量获取
+# 您的其他基础参数，可以自定义增加，但是batch_size请保留，并且设置正确的值
+batch_size=64
+
+if [ x"${etp_flag}" != xtrue ];
+then
+ python3.7 ./main_AttSets.py --data_url=${data_path} --train_url=${output_path}
+else
+ python3.7 ./main_AttSets.py --data_url=${data_path} --train_url=${output_path} > ${print_log}
+fi
+
+# 性能相关数据计算
+# StepTime=`grep "sec/step :" ${print_log} | tail -n 10 | awk '{print $NF}' | awk '{sum+=$1} END {print sum/NR}'`
+# FPS=`awk 'BEGIN{printf "%.2f\n", '${batch_size}'/'${StepTime}'}'`
+
+# 精度相关数据计算
+train_accuracy=`cat ${print_log} | grep "train single ref_iou" | tail -n 200 | awk -F"[" '{print $NF}' | awk -F"]" '{print $1}' | awk '{sum+=$1} END {print sum/NR}'`
+# 提取所有loss打印信息
+grep "train single ref_iou: " ${print_log} | awk -F"[" '{print $NF}' | awk -F"]" '{print $1}' > ./test/output/${ASCEND_DEVICE_ID}/my_output_loss.txt
+
+###########################################################
+#########后面的所有内容请不要修改###########################
+#########后面的所有内容请不要修改###########################
+#########后面的所有内容请不要修改###########################
+###########################################################
+
+# 判断本次执行是否正确使用Ascend NPU
+use_npu_flag=`grep "The model has been compiled on the Ascend AI processor" ${print_log} | wc -l`
+if [ x"${use_npu_flag}" == x0 ];
+then
+ echo "------------------ ERROR NOTICE START ------------------"
+ echo "ERROR, your task haven't used Ascend NPU, please check your npu Migration."
+ echo "------------------ ERROR NOTICE END------------------"
+else
+ echo "------------------ INFO NOTICE START------------------"
+ echo "INFO, your task have used Ascend NPU, please check your result."
+ echo "------------------ INFO NOTICE END------------------"
+fi
+
+# 获取最终的casename，请保留，case文件名为${CaseName}
+get_casename
+
+# 重命名loss文件
+if [ -f ./test/output/${ASCEND_DEVICE_ID}/my_output_loss.txt ];
+then
+ mv ./test/output/${ASCEND_DEVICE_ID}/my_output_loss.txt ./test/output/${ASCEND_DEVICE_ID}/${CaseName}_loss.txt
+fi
+
+# 训练端到端耗时
+end_time=$(date +%s)
+e2e_time=$(( $end_time - $start_time ))
+
+echo "------------------ Final result ------------------"
+# 输出性能FPS/单step耗时/端到端耗时
+echo "Final Performance images/sec : $FPS"
+echo "Final Performance sec/step : $StepTime"
+echo "E2E Training Duration sec : $e2e_time"
+
+# 输出训练精度
+echo "Final Train Accuracy : ${train_accuracy}"
+
+# 最后一个迭代loss值，不需要修改
+ActualLoss=(`awk 'END {print $NF}' $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}_loss.txt`)
+
+#关键信息打印到${CaseName}.log中，不需要修改
+echo "Network = ${Network}" > $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "RankSize = ${RANK_SIZE}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "BatchSize = ${batch_size}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "DeviceType = `uname -m`" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "CaseName = ${CaseName}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "ActualFPS = ${FPS}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "TrainingTime = ${StepTime}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "ActualLoss = ${ActualLoss}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
+echo "E2ETrainingTime = ${e2e_time}" >> $cur_path/output/$ASCEND_DEVICE_ID/${CaseName}.log
\ No newline at end of file
--
Gitee