From 6be7f350a7bdd0f42caca33bc410c6cc7d280d1b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:50:38 +0000
Subject: [PATCH 1/8] =?UTF-8?q?=E6=96=B0=E5=BB=BA=20IEG=5FID2049=5Ffor=5FA?=
=?UTF-8?q?CL?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep | 0
1 file changed, 0 insertions(+), 0 deletions(-)
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep
new file mode 100644
index 000000000..e69de29bb
--
Gitee
From 90d12a7ebd34bcb74f1415558f730c0721572f82 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:51:27 +0000
Subject: [PATCH 2/8] readme
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Signed-off-by: 虽千万人吾往矣 <1291632293@qq.com>
---
.../contrib/cv/IEG_ID2049_for_ACL/README.md | 82 +++++++++++++++++++
1 file changed, 82 insertions(+)
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
new file mode 100644
index 000000000..9029e469d
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
@@ -0,0 +1,82 @@
+概述
+
+Attention-OCR是一个基于卷积神经网络CNN、循环神经网络RNN以及一种新颖的注意机制的自然场景文本检测识别网络。
+
+本文提出了一个高效率的噪声标签训练方法。
+
+
+- 参考论文：
+[Distilling Effective Supervision from Severe Label Noise](https://arxiv.org/pdf/1910.00701.pdf),
+ CVPR2020
+- 参考实现：
+https://github.com/google-research/google-research/tree/master/ieg
+
+- 适配昇腾 AI 处理器的实现：
+https://gitee.com/lwrstudy/ModelZoo-TensorFlow/tree/master/TensorFlow/contrib/cv/Slot-Attention_ID2028_for_TensorFlow
+
+
+原始模型
+
+obs地址：obs://lwr-npu/tl/checkp
+
+
+步骤一:将checkpoint.ckpt-85329转化成IEG.pb
+通过代码freeze_graph将ckpt转成pb
+
+
+
+pb模型
+
+```
+IEG.pb
+```
+obs地址：obs://lwr-npu/tl/IEG.pb
+
+
+om模型
+
+转IEG.pb到hpiegmodel.om
+
+使用ATC模型转换工具进行模型转换时可以参考如下指令:
+
+```
+atc --model=./IEG.pb --input_shape=“input:100,3,32,32” --framework=3 --output=hpiegmodel --soc_version=Ascend910A --precision_mode=force_fp32 --op_select_implmode=high_precision
+```
+
+成功转化成hpiegmodel.om
+
+hpiegmodel.om的obs地址：obs://lwr-npu/tl/hpiegmodel.om
+
+
+数据集转换bin
+1.使用convertimg.py将数据集转为png，代码中的src为数据集源路径，dest为png数据集输出路径
+2.参考 https://gitee.com/ascend/tools/tree/master/img2bin, 获取msame推理工具及使用方法。
+这里对该代码库中的img2bin进行了少量修改，加入了数据预处理
+使用img2bin.py将png数据集转为bin格式，img2bin.py为tools-master中的代码文件,
+```
+python3 img2bin.py -i /mnt/home/test_user01/toimg -w 32 -h 32 -f RGB -a NHWC -t float32 -c [1,1,1] -o ./imgout
+```
+3.使用merge.py将bin格式的数据集合并为一个batch
+转换后的bin文件见obs地址:obs://lwr-npu/tl/newinputbin
+
+
+
+
+使用msame工具推理
+
+参考 https://gitee.com/ascend/tools/tree/master/msame, 获取msame推理工具及使用方法。
+
+使用msame推理工具，参考如下命令，发起推理测试：
+
+```
+./msame --model "/mnt/home/test_user01/hpiegmodel.om" --input "/home/test_user01/slottl/tools-master/img2bin/newinput.bin" --output "./iegout/" --outfmt TXT --loop 1
+```
+
+
+
+
+
+
+
+
+
--
Gitee
From 9b81ba917af3071fca4fdfbd7d64170315fffa2b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:52:07 +0000
Subject: [PATCH 3/8] update
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Signed-off-by: 虽千万人吾往矣 <1291632293@qq.com>
---
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md | 3 ---
1 file changed, 3 deletions(-)
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
index 9029e469d..fe293069f 100644
--- a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
@@ -1,7 +1,4 @@
概述
-
-Attention-OCR是一个基于卷积神经网络CNN、循环神经网络RNN以及一种新颖的注意机制的自然场景文本检测识别网络。
-
本文提出了一个高效率的噪声标签训练方法。
--
Gitee
From 29d6a7eba03801145efa8a684039afc1f11e35a8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:53:37 +0000
Subject: [PATCH 4/8] update
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Signed-off-by: 虽千万人吾往矣 <1291632293@qq.com>
---
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
index fe293069f..6559e5cdb 100644
--- a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
@@ -9,7 +9,7 @@
https://github.com/google-research/google-research/tree/master/ieg
- 适配昇腾 AI 处理器的实现：
-https://gitee.com/lwrstudy/ModelZoo-TensorFlow/tree/master/TensorFlow/contrib/cv/Slot-Attention_ID2028_for_TensorFlow
+https://gitee.com/lwrstudy/ModelZoo-TensorFlow/tree/master/TensorFlow/contrib/cv/IEG_ID2049_for_TensorFlow
原始模型
--
Gitee
From 0e0dc8096494be71a52de137b06b0755146dd4c5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:54:53 +0000
Subject: [PATCH 5/8] =?UTF-8?q?=E4=BB=A3=E7=A0=81=E6=96=87=E4=BB=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Signed-off-by: 虽千万人吾往矣 <1291632293@qq.com>
---
.../cv/IEG_ID2049_for_ACL/convertimg.py | 97 ++++++
.../contrib/cv/IEG_ID2049_for_ACL/copyimg.py | 6 +
.../cv/IEG_ID2049_for_ACL/frozen_graph.py | 58 ++++
.../contrib/cv/IEG_ID2049_for_ACL/img2bin.py | 310 ++++++++++++++++++
.../contrib/cv/IEG_ID2049_for_ACL/merge.py | 16 +
5 files changed, 487 insertions(+)
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/convertimg.py
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/copyimg.py
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/frozen_graph.py
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/img2bin.py
create mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/merge.py
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/convertimg.py b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/convertimg.py
new file mode 100644
index 000000000..3ab59bf81
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/convertimg.py
@@ -0,0 +1,97 @@
+import numpy as np
+import os
+from npu_bridge.npu_init import *
+
+import collections
+import math
+import sys
+
+from absl import flags
+from PIL import Image
+from six.moves import cPickle
+import sklearn.metrics as sklearn_metrics
+import tensorflow.compat.v1 as tf
+
+def load_batch(fpath, label_key='labels'):
+ """Internal utility for parsing CIFAR data.
+
+ Args:
+ fpath: path the file to parse.
+ label_key: key for label data in the retrieve dictionary.
+
+ Returns:
+ A tuple `(data, labels)`.
+ """
+
+ with tf.io.gfile.GFile(fpath, 'rb') as f:
+ if sys.version_info < (3,):
+ d = cPickle.load(f)
+ else:
+ d = cPickle.load(f, encoding='bytes')
+ # decode utf8
+ d_decoded = {}
+ for k, v in d.items():
+ d_decoded[k.decode('utf8')] = v
+ d = d_decoded
+ data = d['data']
+ labels = d[label_key]
+
+ data = data.reshape(data.shape[0], 3, 32, 32)
+ return data, labels
+
+def cifar10_load_data(root):
+ """Loads CIFAR10 dataset.
+
+ Args:
+ root: path that saves data file.
+
+ Returns:
+ Tuple of Numpy arrays: `(x_train, y_train), (x_test, y_test)`.
+ """
+ if not root:
+ return tf.keras.datasets.cifar10.load_data()
+
+ num_train_samples = 50000
+
+ x_train = np.empty((num_train_samples, 3, 32, 32), dtype='uint8')
+ y_train = np.empty((num_train_samples,), dtype='uint8')
+
+ for i in range(1, 6):
+ fpath = os.path.join(root, 'data_batch_' + str(i))
+ (x_train[(i - 1) * 10000:i * 10000, :, :, :],
+ y_train[(i - 1) * 10000:i * 10000]) = load_batch(fpath)
+
+ fpath = os.path.join(root, 'test_batch')
+ x_test, y_test = load_batch(fpath)
+
+ y_train = np.reshape(y_train, (len(y_train), 1))
+ y_test = np.reshape(y_test, (len(y_test), 1))
+
+ x_train = x_train.transpose(0, 2, 3, 1)
+ x_test = x_test.transpose(0, 2, 3, 1)
+
+ x_test = x_test.astype(x_train.dtype)
+ y_test = y_test.astype(y_train.dtype)
+
+ return (x_train, y_train), (x_test, y_test)
+
+
+
+
+
+def convertimg(src,dest):
+ (x_train, y_train), (x_test,y_test)=cifar10_load_data(src)
+ print("xtest的shape",str(x_test.shape))
+ print("xtrain的shape", str(x_train.shape))
+ print("ytest的shape",str(y_test.shape))
+ print("ytrain的shape", str(y_train.shape))
+
+ for i in range(100):
+ im = Image.fromarray(x_test[i])
+ im.save(dest+"/cifar10_test_"+str(i)+".png")
+
+
+if __name__ == '__main__':
+ src="./data/cifar-10-python/cifar-10-batches-py"
+ dest="./toimg"
+ convertimg(src=src,dest=dest)
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/copyimg.py b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/copyimg.py
new file mode 100644
index 000000000..0c61aa740
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/copyimg.py
@@ -0,0 +1,6 @@
+import os
+import shutil
+filelist=os.listdir("/home/test_user01/slottl/tools-master/img2bin/imgout/")
+for i in range(100):
+ filepath=os.path.join("/home/test_user01/slottl/tools-master/img2bin/imgout/", filelist[i])
+ shutil.copy(filepath,"./partimg")
\ No newline at end of file
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/frozen_graph.py b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/frozen_graph.py
new file mode 100644
index 000000000..7849b887f
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/frozen_graph.py
@@ -0,0 +1,58 @@
+# Copyright 2020 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 tensorflow as tf
+from tensorflow.python.tools import freeze_graph
+from tensorflow.python.framework import graph_util
+from nets import nets_factory
+import argparse
+from models.basemodel import create_network
+def parse_args():
+ parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+ parser.add_argument('--ckpt_path', default="",
+ help="""set checkpoint path""")
+ args, unknown_args = parser.parse_known_args()
+ if len(unknown_args) > 0:
+ for bad_arg in unknown_args:
+ print("ERROR: Unknown command line arg: %s" % bad_arg)
+ raise ValueError("Invalid command line arg(s)")
+ return args
+
+def main():
+ args = parse_args()
+ tf.reset_default_graph()
+ # set inputs node
+ inputs = tf.placeholder(tf.float32, shape=[None, 224, 224, 3], name="input")
+ # create inference graph
+ # network_fn = nets_factory.get_network_fn('mobilenet_v1', num_classes=1001, weight_decay=0.0, is_training=False)
+ # logits, end_points = network_fn(inputs, reuse=tf.AUTO_REUSE)
+ net=create_network("wrn28-10", 10)
+ logits=net(inputs, name='model', reuse=tf.AUTO_REUSE, training=True)
+ with tf.Session() as sess:
+ tf.train.write_graph(sess.graph_def, './', 'model.pb')
+ freeze_graph.freeze_graph(
+ input_graph='model.pb',
+ input_saver='',
+ input_binary=False,
+ input_checkpoint=args.ckpt_path,
+ output_node_names='iegoutput', # graph outputs node
+ restore_op_name='save/restore_all',
+ filename_tensor_name='save/Const:0',
+ output_graph='IEG.pb', # graph outputs name
+ clear_devices=False,
+ initializer_nodes='')
+ print("done")
+
+if __name__ == '__main__':
+ main()
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/img2bin.py b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/img2bin.py
new file mode 100644
index 000000000..b78efdd8f
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/img2bin.py
@@ -0,0 +1,310 @@
+# -*- coding: UTF-8 -*-
+#
+# =======================================================================
+#
+# Copyright (C) 2018, Hisilicon Technologies Co., Ltd. All Rights Reserved.
+#
+# 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.
+#
+# 3 Neither the names of the copyright holders nor the names of the
+# contributors may be used to endorse or promote products derived from this
+# software without specific prior written permission.
+#
+# 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.
+# =======================================================================
+from __future__ import print_function
+import argparse
+import os
+import sys
+import json
+# import getpass
+
+IMG_EXT = ['.jpg', '.JPG', '.png', '.PNG', '.bmp', '.BMP', '.jpeg', '.JPEG']
+
+try:
+ import cv2 as cv
+except ImportError:
+ if sys.version_info.major == 2:
+ confirm = raw_input("[info] Begin to install opencv-python, input[Y/N]:")
+ else:
+ confirm = input("[info] Begin to install opencv-python, input[Y/N]:")
+
+ if confirm == 'Y' or confirm == 'y':
+ print('[info] Starting to install opencv-python...')
+ if sys.version_info.major == 2:
+ import commands
+ retu, output = commands.getstatusoutput("sudo yum install -y opencv-python")
+ if retu != 0:
+ retu, output = commands.getstatusoutput("sudo apt-get install -y python-opencv")
+ if retu != 0:
+ print('[ERROR] install opencv-python failed,please check env.')
+ exit(0)
+ else:
+ retu = os.system('sudo python3 -m pip install opencv-python')
+ if retu != 0:
+ print('[ERROR] install opencv-python failed,please check env.')
+ exit(0)
+
+ import cv2 as cv
+ else:
+ print("[info] The installation has been cancled.")
+ exit(0)
+
+try:
+ import numpy as np
+except ImportError:
+ if sys.version_info.major == 2:
+ os.system('pip2 install numpy')
+ else:
+ os.system('python3 -m pip install numpy')
+ import numpy as np
+
+def get_args():
+ parser = argparse.ArgumentParser(
+ conflict_handler='resolve',
+ description='''eg1: python2 img2bin.py
+ -i ./images -w 416 -h 416 -f BGR -a NHWC -t uint8 -m [104,117,123] -c [1,1,1] -o ./out
+ eg2: python2 img2bin.py -i ./test.txt -t uint8 -o ./out''')
+ parser.add_argument('-i', '--input', required=True, type=str, \
+ help='folder of input image or file of other input.')
+ parser.add_argument('-w', '--width', required=True, type=int, \
+ help='resized image width before inference.')
+ parser.add_argument('-h', '--height', required=True, type=int, \
+ help='resized image height before inference.')
+ parser.add_argument('-f', '--output_image_format', default='BGR', type=str, \
+ help='output image format in (BGR/RGB/YUV/GRAY).')
+ parser.add_argument('-a', '--output_format', default='NCHW', type=str, \
+ help='output format in (NCHW/NHWC).')
+ parser.add_argument('-t', '--output_type', required=True, type=str, \
+ help='output type in (float32/uint8/int32/uint32).')
+ parser.add_argument('-m', '--mean', default='[0, 0, 0]', \
+ help='reduce mean for each channel.')
+ parser.add_argument('-c', '--coefficient', default='[1, 1, 1]', \
+ help='multiplying coefficients for each channel.')
+ parser.add_argument('-o', '--output', default='./', \
+ help='output path.')
+ return parser.parse_args()
+
+
+def eprint(*args, **kwargs):
+ """print error message to stderr
+ """
+ print(*args, file=sys.stderr, **kwargs)
+
+
+def check_args(args):
+ """
+ check console parameters according to restrictions.
+ return: True or False
+ """
+ check_flag = True
+ is_dir = True
+ if os.path.isdir(args.input):
+ if not os.listdir(args.input):
+ eprint('[ERROR] input image path=%r is empty.' % args.input)
+ check_flag = False
+ elif os.path.isfile(args.input):
+ is_dir = False
+ else:
+ eprint('[ERROR] input path=%r does not exist.' % args.input)
+ check_flag = False
+ if args.output_image_format not in ('BGR','RGB', 'YUV', 'GRAY'):
+ eprint("[ERROR] Convert to %d is not support." % (args.output_image_format))
+ check_flag = False
+ if args.height <= 0 or args.width <= 0:
+ eprint("[ERROR] image height or image width must be greater than 0.")
+ check_flag = False
+ elif args.output_image_format == 'YUV':
+ if args.width % 2 == 1 or args.height % 2 == 1:
+ eprint("[ERROR] when the output color format is YUV, the width and height of the picture must be even.")
+ check_flag = False
+ return check_flag, is_dir
+
+
+def convert_img_2_yuv(input_img):
+ input_img_height = input_img.shape[0]
+ input_img_width = input_img.shape[1]
+ bgr2y_list = np.array([29, 150, 77])
+ bgr2y_data = input_img.reshape(input_img_height * input_img_width, 3)
+ y_data = np.dot(bgr2y_data, bgr2y_list) >> 8
+ bgr2u_list = np.array([131, -87, -44])
+ bgr2v_list = np.array([-21, -110, 131])
+ bgr2uv_matrix = np.transpose(np.append(bgr2u_list, bgr2v_list).reshape((2, 3)))
+ bgr2uv_data = input_img[0::2, 0::2, :].reshape((input_img_height // 2 * input_img_width // 2, 3))
+ yuv_base_data = np.dot(bgr2uv_data, bgr2uv_matrix) >> 8
+ u_data = yuv_base_data[:,0] + 128
+ v_data = yuv_base_data[:,1] + 128
+ u_v_data = np.transpose(np.append(u_data.flatten(), v_data.flatten()).reshape((2, input_img_height //2 * input_img_width // 2)))
+ nv12_data = np.append(y_data.flatten(), u_v_data.flatten()).reshape((input_img_height // 2 * 3, input_img_width)).astype(np.uint8)
+ return nv12_data
+
+
+def convert_img(args, input_img):
+ if args.output_image_format == 'BGR':
+ converted_input_img = input_img
+ elif args.output_image_format == 'RGB':
+ converted_input_img = cv.cvtColor(input_img, cv.COLOR_BGR2RGB)
+ elif args.output_image_format == 'YUV':
+ converted_input_img = convert_img_2_yuv(input_img)
+ elif args.output_image_format == 'GRAY':
+ converted_input_img = cv.cvtColor(input_img, cv.COLOR_BGR2GRAY)
+ return converted_input_img
+
+
+def resize_img(args, input_img):
+ old_size = input_img.shape[0:2]
+ target_size = [args.height, args.width]
+ ratio = min(float(target_size[i]) / (old_size[i]) for i in range(len(old_size)))
+ new_size = tuple([int(i*ratio) for i in old_size])
+ img_new = cv.resize(input_img,(new_size[1], new_size[0]))
+ pad_w = target_size[1] - new_size[1]
+ pad_h = target_size[0] - new_size[0]
+ top, bottom = pad_h // 2, pad_h - (pad_h // 2)
+ left, right = pad_w // 2, pad_w - (pad_w // 2)
+ resized_img = cv.copyMakeBorder(img_new, top, bottom, left, right, cv.BORDER_CONSTANT, None,(0, 0, 0))
+ return resized_img
+
+
+def change_type(args, input_img):
+ if args.output_type == 'float32':
+ change_type_img = input_img.astype(np.float32)
+ elif args.output_type == 'int32':
+ change_type_img = input_img.astype(np.int32)
+ elif args.output_type == 'uint32':
+ change_type_img = input_img.astype(np.uint32)
+ else:
+ change_type_img = input_img.astype(np.uint8)
+ return change_type_img
+
+
+def mean(args, input_img):
+ if isinstance (args.mean, str):
+ args.mean = json.loads(args.mean)
+ input_img = input_img.astype(np.float32)
+ if args.output_image_format == 'GRAY':
+ input_img[:, :] -= args.mean[0]
+ elif args.output_image_format in ('BGR', 'RGB'):
+ input_img[:, :, 0] -= args.mean[0]
+ input_img[:, :, 1] -= args.mean[1]
+ input_img[:, :, 2] -= args.mean[2]
+ return input_img
+
+
+def coefficient(args, input_img):
+ """
+ Normalize the input image
+ """
+ if isinstance(args.coefficient, str):
+ args.coefficient = json.loads(args.coefficient)
+ input_img = input_img.astype(np.float32)
+ if args.output_image_format == 'GRAY':
+ input_img[:, :] *= args.coefficient[0]
+ elif args.output_image_format in ('BGR', 'RGB'):
+ input_img[:, :, 0] *= args.coefficient[0]
+ input_img[:, :, 1] *= args.coefficient[1]
+ input_img[:, :, 2] *= args.coefficient[2]
+ print("------执行ieg的处理--------")
+ print(input_img)
+ input_img=input_img/255.0
+ input_img=input_img-0.5
+ input_img=input_img*2
+ print("--------执行之后的结果---------")
+ print(input_img)
+ return input_img
+
+
+def change_format(args, input_img):
+ if args.output_format == 'NCHW':
+ if args.output_image_format in ('RGB', 'BGR'):
+ change_format_img = input_img.transpose(2,0,1).copy()
+ return change_format_img
+ return input_img
+
+
+def mkdir_output(args):
+ if not os.path.exists(args.output):
+ os.makedirs(args.output)
+ return
+
+
+def process(args, file_path):
+ if file_path.endswith(".txt"):
+ if sys.version_info.major == 2:
+ import ConfigParser as configparser
+ else:
+ import configparser
+
+ config = configparser.ConfigParser()
+ config.read(file_path)
+ if sys.version_info.major == 2:
+ input_node = json.loads(config.get('baseconf', 'input_node'))
+ shape = json.loads(config.get('baseconf', 'shape'))
+ else:
+ input_node = json.loads(config['baseconf']['input_node'])
+ shape = json.loads(config['baseconf']['shape'])
+ input_node_np = np.array(input_node)
+ change_type_img_info = change_type(args, input_node_np)
+ img_info = np.reshape(change_type_img_info, shape)
+ out_path = os.path.join(args.output, os.path.splitext(os.path.split(file_path)[1])[0] + ".bin")
+ mkdir_output(args)
+ img_info.tofile(out_path)
+ else:
+ input_img = cv.imread(file_path)
+ resized_img1 = resize_img(args, input_img)
+ converted_img = convert_img(args, resized_img1)
+ if args.output_image_format == "YUV":
+ change_format_img = converted_img
+ else:
+ mean_img = mean(args, converted_img)
+ coefficient_img = coefficient(args, mean_img)
+ change_type_img = change_type(args, coefficient_img)
+ change_format_img = change_format(args, change_type_img)
+ out_path = os.path.join(args.output, os.path.splitext(os.path.split(file_path)[1])[0] + ".bin")
+ mkdir_output(args)
+ change_format_img.tofile(out_path)
+
+
+def main():
+ """main function to receive params them change data to bin.
+ """
+ args = get_args()
+ ret, is_dir = check_args(args)
+ if ret:
+ if is_dir:
+ files_name = os.listdir(args.input)
+ for file_name in files_name:
+ if os.path.splitext(file_name)[1] in IMG_EXT:
+ file_path = os.path.join(args.input, file_name)
+ process(args, file_path)
+ print("[info] bin file generated successfully.")
+ else:
+ if os.path.splitext(args.input)[1] in IMG_EXT or os.path.splitext(args.input)[1] == "txt":
+ process(args, args.input)
+ print("[info] bin file generated successfully.")
+ else:
+ eprint("[ERROR] input file must be image or end with '.txt'.")
+
+if __name__ == '__main__':
+ main()
+
+
+
+
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/merge.py b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/merge.py
new file mode 100644
index 000000000..4c5d39f46
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/merge.py
@@ -0,0 +1,16 @@
+import os
+import numpy as np
+import shutil
+
+
+data_sets = []
+sample_batch_input_bin_dir = "./imgout/"
+for item in os.listdir(sample_batch_input_bin_dir):
+ # 读取bin文件时，bin文件内的dtype类型须根据模型的输入类型确定，下面以float32为例
+ original_input_data = np.fromfile(os.path.join(sample_batch_input_bin_dir, item), dtype=np.int32)
+ # 将数据重新组织，具体根据模型输入中的shape值确定
+ current_input_data = original_input_data.reshape(32, 32, 3)
+ # 将当前的数据添加到列表中
+ data_sets.append(current_input_data)
+# 将每个batch的数据保存到一个输入bin文件中，从而得到一个包含多batch的输入bin文件
+np.array(data_sets).tofile("newinput.bin")
\ No newline at end of file
--
Gitee
From ff57b6e351cc0a8fb244f572a2c5ac14ff038860 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:55:03 +0000
Subject: [PATCH 6/8] =?UTF-8?q?=E5=88=A0=E9=99=A4=E6=96=87=E4=BB=B6=20ACL?=
=?UTF-8?q?=5FTensorFlow/contrib/cv/IEG=5FID2049=5Ffor=5FACL/.keep?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep | 0
1 file changed, 0 insertions(+), 0 deletions(-)
delete mode 100644 ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/.keep
deleted file mode 100644
index e69de29bb..000000000
--
Gitee
From 64dc62259ea8ae940ead3cebd3255b15f018ecf3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:57:24 +0000
Subject: [PATCH 7/8] update
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Signed-off-by: 虽千万人吾往矣 <1291632293@qq.com>
---
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
index 6559e5cdb..64792b2ba 100644
--- a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
@@ -47,12 +47,16 @@ hpiegmodel.om的obs地址：obs://lwr-npu/tl/hpiegmodel.om
数据集转换bin
1.使用convertimg.py将数据集转为png，代码中的src为数据集源路径，dest为png数据集输出路径
+
2.参考 https://gitee.com/ascend/tools/tree/master/img2bin, 获取msame推理工具及使用方法。
+
这里对该代码库中的img2bin进行了少量修改，加入了数据预处理
-使用img2bin.py将png数据集转为bin格式，img2bin.py为tools-master中的代码文件,
+使用img2bin.py将png数据集转为bin格式，img2bin.py为tools-master中的代码文件
+
```
python3 img2bin.py -i /mnt/home/test_user01/toimg -w 32 -h 32 -f RGB -a NHWC -t float32 -c [1,1,1] -o ./imgout
```
+
3.使用merge.py将bin格式的数据集合并为一个batch
转换后的bin文件见obs地址:obs://lwr-npu/tl/newinputbin
--
Gitee
From 26d0f5e156291e380f3c7659758ccfab58ea6e1b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=99=BD=E5=8D=83=E4=B8=87=E4=BA=BA=E5=90=BE=E5=BE=80?=
=?UTF-8?q?=E7=9F=A3?= <1291632293@qq.com>
Date: Fri, 28 Oct 2022 05:58:28 +0000
Subject: [PATCH 8/8] update
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md.
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Signed-off-by: 虽千万人吾往矣 <1291632293@qq.com>
---
ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
index 64792b2ba..e018a9c9d 100644
--- a/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
+++ b/ACL_TensorFlow/contrib/cv/IEG_ID2049_for_ACL/README.md
@@ -48,7 +48,7 @@ hpiegmodel.om的obs地址：obs://lwr-npu/tl/hpiegmodel.om
数据集转换bin
1.使用convertimg.py将数据集转为png，代码中的src为数据集源路径，dest为png数据集输出路径
-2.参考 https://gitee.com/ascend/tools/tree/master/img2bin, 获取msame推理工具及使用方法。
+2.参考 https://gitee.com/ascend/tools/tree/master/img2bin, 获取img2bin推理工具及使用方法。
这里对该代码库中的img2bin进行了少量修改，加入了数据预处理
使用img2bin.py将png数据集转为bin格式，img2bin.py为tools-master中的代码文件
--
Gitee