diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/PointRend.patch" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/PointRend.patch"
deleted file mode 100644
index 50a209c738563810911c149705fd87cb0300e6f6..0000000000000000000000000000000000000000
--- "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/PointRend.patch"
+++ /dev/null
@@ -1,171 +0,0 @@
-1.从mmcv复制替换掉grid_sample的方法
-2.去除部分动态shape的where算子
-3.expand输入类型转换
-4.div输入类型转换
-diff --git a/detectron2/modeling/meta_arch/semantic_seg.py b/detectron2/modeling/meta_arch/semantic_seg.py
-index 7db8410..8a08e76 100644
---- a/detectron2/modeling/meta_arch/semantic_seg.py
-+++ b/detectron2/modeling/meta_arch/semantic_seg.py
-@@ -92,10 +92,9 @@ class SemanticSegmentor(nn.Module):
-               The prediction has shape KxHxW that represents the logits of
-               each class for each pixel.
-         """
--        images = [x["image"].to(self.device) for x in batched_inputs]
-+        '''images = [x["image"].to(self.device) for x in batched_inputs]
-         images = [(x - self.pixel_mean) / self.pixel_std for x in images]
-         images = ImageList.from_tensors(images, self.backbone.size_divisibility)
--
-         features = self.backbone(images.tensor)
- 
-         if "sem_seg" in batched_inputs[0]:
-@@ -104,8 +103,13 @@ class SemanticSegmentor(nn.Module):
-                 targets, self.backbone.size_divisibility, self.sem_seg_head.ignore_value
-             ).tensor
-         else:
--            targets = None
-+            targets = None'''
-+        batched_inputs = batched_inputs.to(self.device)
-+        raw_input = batched_inputs
-+        features = self.backbone(raw_input)
-+        targets = None
-         results, losses = self.sem_seg_head(features, targets)
-+        return results
- 
-         if self.training:
-             return losses
-diff --git a/projects/PointRend/point_rend/point_features.py b/projects/PointRend/point_rend/point_features.py
-index 26e706e..9d1ab3b 100644
---- a/projects/PointRend/point_rend/point_features.py
-+++ b/projects/PointRend/point_rend/point_features.py
-@@ -25,6 +25,88 @@ def _as_tensor(x):
-     return torch.as_tensor(x)
- 
- 
-+def bilinear_grid_sample(im, grid, align_corners=False):
-+    """Given an input and a flow-field grid, computes the output using input
-+    values and pixel locations from grid. Supported only bilinear interpolation
-+    method to sample the input pixels.
-+
-+    Args:
-+        im (torch.Tensor): Input feature map, shape (N, C, H, W)
-+        grid (torch.Tensor): Point coordinates, shape (N, Hg, Wg, 2)
-+        align_corners {bool}: If set to True, the extrema (-1 and 1) are
-+            considered as referring to the center points of the input's
-+            corner pixels. If set to False, they are instead considered as
-+            referring to the corner points of the input's corner pixels,
-+            making the sampling more resolution agnostic.
-+    Returns:
-+        torch.Tensor: A tensor with sampled points, shape (N, C, Hg, Wg)
-+    """
-+    n, c, h, w = im.shape
-+    gn, gh, gw, _ = grid.shape
-+    assert n == gn
-+
-+    x = grid[:, :, :, 0]
-+    y = grid[:, :, :, 1]
-+
-+    if align_corners:
-+        x = ((x + 1) / 2) * (w - 1)
-+        y = ((y + 1) / 2) * (h - 1)
-+    else:
-+        x = ((x + 1) * w - 1) / 2
-+        y = ((y + 1) * h - 1) / 2
-+
-+    x = x.view(n, -1)
-+    y = y.view(n, -1)
-+
-+    x0 = torch.floor(x).long()
-+    y0 = torch.floor(y).long()
-+    x1 = x0 + 1
-+    y1 = y0 + 1
-+
-+    wa = ((x1 - x) * (y1 - y)).unsqueeze(1)
-+    wb = ((x1 - x) * (y - y0)).unsqueeze(1)
-+    wc = ((x - x0) * (y1 - y)).unsqueeze(1)
-+    wd = ((x - x0) * (y - y0)).unsqueeze(1)
-+
-+    # Apply default for grid_sample function zero padding
-+    im_padded = F.pad(im, pad=[1, 1, 1, 1], mode='constant', value=0)
-+    padded_h = h + 2
-+    padded_w = w + 2
-+    # save points positions after padding
-+    x0, x1, y0, y1 = x0 + 1, x1 + 1, y0 + 1, y1 + 1
-+
-+    # Clip coordinates to padded image size
-+    x0 = torch.where(x0 < 0, torch.tensor(0), x0)
-+    x0 = torch.where(x0 > padded_w - 1, torch.tensor(padded_w - 1), x0)
-+    x1 = torch.where(x1 < 0, torch.tensor(0), x1)
-+    x1 = torch.where(x1 > padded_w - 1, torch.tensor(padded_w - 1), x1)
-+    y0 = torch.where(y0 < 0, torch.tensor(0), y0)
-+    y0 = torch.where(y0 > padded_h - 1, torch.tensor(padded_h - 1), y0)
-+    y1 = torch.where(y1 < 0, torch.tensor(0), y1)
-+    y1 = torch.where(y1 > padded_h - 1, torch.tensor(padded_h - 1), y1)
-+
-+    im_padded = im_padded.view(n, c, -1)
-+
-+    x0_y0_tmp = (x0 + y0 * padded_w).unsqueeze(1).to(dtype=torch.int32)
-+    s = torch.zeros(x0_y0_tmp.size(0), c, x0_y0_tmp.size(2))
-+    x0_y0 = x0_y0_tmp.expand_as(s).to(dtype=torch.int64)
-+    x0_y1_tmp = (x0 + y1 * padded_w).unsqueeze(1).to(dtype=torch.int32)
-+    s = torch.zeros(x0_y1_tmp.size(0), c, x0_y1_tmp.size(2))
-+    x0_y1 = x0_y1_tmp.expand_as(s).to(dtype=torch.int64)
-+    x1_y0_tmp = (x1 + y0 * padded_w).unsqueeze(1).to(dtype=torch.int32)
-+    s = torch.zeros(x1_y0_tmp.size(0), c, x1_y0_tmp.size(2))
-+    x1_y0 = x1_y0_tmp.expand_as(s).to(dtype=torch.int64)
-+    x1_y1_tmp = (x1 + y1 * padded_w).unsqueeze(1).to(dtype=torch.int32)
-+    s = torch.zeros(x1_y1_tmp.size(0), c, x1_y1_tmp.size(2))
-+    x1_y1 = x1_y1_tmp.expand_as(s).to(dtype=torch.int64)
-+
-+    Ia = torch.gather(im_padded, 2, x0_y0)
-+    Ib = torch.gather(im_padded, 2, x0_y1)
-+    Ic = torch.gather(im_padded, 2, x1_y0)
-+    Id = torch.gather(im_padded, 2, x1_y1)
-+
-+    return (Ia * wa + Ib * wb + Ic * wc + Id * wd).reshape(n, c, gh, gw)
-+
- def point_sample(input, point_coords, **kwargs):
-     """
-     A wrapper around :function:`torch.nn.functional.grid_sample` to support 3D point_coords tensors.
-@@ -44,8 +126,11 @@ def point_sample(input, point_coords, **kwargs):
-     add_dim = False
-     if point_coords.dim() == 3:
-         add_dim = True
--        point_coords = point_coords.unsqueeze(2)
--    output = F.grid_sample(input, 2.0 * point_coords - 1.0, **kwargs)
-+        s = torch.zeros(point_coords.size(0), point_coords.size(1), 1, point_coords.size(2))
-+        point_coords = point_coords.reshape(point_coords.size(0), point_coords.size(1), 1, point_coords.size(2))
-+    #output = F.grid_sample(input, 2.0 * point_coords - 1.0, **kwargs)
-+    # copy from mmcv
-+    output = bilinear_grid_sample(input, 2.0 * point_coords - 1.0, align_corners=False)
-     if add_dim:
-         output = output.squeeze(3)
-     return output
-@@ -147,8 +232,11 @@ def get_uncertain_point_coords_on_grid(uncertainty_map, num_points):
-     num_points = min(H * W, num_points)
-     point_indices = torch.topk(uncertainty_map.view(R, H * W), k=num_points, dim=1)[1]
-     point_coords = torch.zeros(R, num_points, 2, dtype=torch.float, device=uncertainty_map.device)
--    point_coords[:, :, 0] = w_step / 2.0 + (point_indices % W).to(torch.float) * w_step
--    point_coords[:, :, 1] = h_step / 2.0 + (point_indices // W).to(torch.float) * h_step
-+    point_indices = point_indices.to(torch.int32)
-+    x = w_step / 2.0 + (point_indices % W).to(torch.float) * w_step
-+    y = h_step / 2.0 + (point_indices // W).to(torch.float) * h_step
-+    point_coords = torch.stack([x, y], dim=-1)
-+    point_indices = point_indices.to(torch.int64)
-     return point_indices, point_coords
- 
- 
-diff --git a/projects/PointRend/point_rend/semantic_seg.py b/projects/PointRend/point_rend/semantic_seg.py
-index ea65200..ba5d552 100644
---- a/projects/PointRend/point_rend/semantic_seg.py
-+++ b/projects/PointRend/point_rend/semantic_seg.py
-@@ -126,7 +126,9 @@ class PointRendSemSegHead(nn.Module):
- 
-                 # put sem seg point predictions to the right places on the upsampled grid.
-                 N, C, H, W = sem_seg_logits.shape
--                point_indices = point_indices.unsqueeze(1).expand(-1, C, -1)
-+                point_indices = point_indices.unsqueeze(1)
-+                s = torch.zeros(point_indices.size(0), C, point_indices.size(2))
-+                point_indices = point_indices.to(dtype=torch.int32).expand_as(s).to(dtype=torch.int64)
-                 sem_seg_logits = (
-                     sem_seg_logits.reshape(N, C, H * W)
-                     .scatter_(2, point_indices, point_logits)
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/PointRend_pth2onnx.py" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/PointRend_pth2onnx.py"
deleted file mode 100644
index f6588e243ebe31f590e53f7e75a0e70c095900e1..0000000000000000000000000000000000000000
--- "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/PointRend_pth2onnx.py"
+++ /dev/null
@@ -1,25 +0,0 @@
-import os
-import torch
-from detectron2.config import get_cfg
-from detectron2.projects.point_rend import add_pointrend_config
-from detectron2.engine import DefaultTrainer
-from detectron2.checkpoint import DetectionCheckpointer
-
-def setup():
-    cfg = get_cfg()
-    add_pointrend_config(cfg)
-    cfg.merge_from_file(os.getcwd() + '/projects/PointRend/configs/SemanticSegmentation/pointrend_semantic_R_101_FPN_1x_cityscapes.yaml')
-    cfg.MODEL.DEVICE = 'cpu'
-    cfg.freeze()
-    return cfg
-
-cfg = setup()
-model = DefaultTrainer.build_model(cfg)
-DetectionCheckpointer(model, save_dir=cfg.OUTPUT_DIR).resume_or_load('./model_final_cf6ac1.pkl')
-model.eval()
-
-input_name = ['images']
-output_name = ['results']
-dynamic_axes = {'images': {0: '-1'}, 'results': {0: '-1'}}
-dummy_input = torch.randn(1, 3, 1024, 2048)
-torch.onnx.export(model, dummy_input, 'pointrend.onnx', input_names=input_name, dynamic_axes=dynamic_axes, output_names=output_name, opset_version=11, verbose=True)
\ No newline at end of file
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/RefineDet.PyTorch" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/RefineDet.PyTorch"
deleted file mode 100644
index b8a24babb6ed49932c891499d7d8d47a4e9826bf..0000000000000000000000000000000000000000
--- "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/ONNX\346\250\241\345\236\213\346\216\250\347\220\206\346\214\207\345\257\274/benchmark/cv/segmentation/RefineDet.PyTorch"
+++ /dev/null
@@ -1 +0,0 @@
-直接导出onnx使用onnxsim后为空图
\ No newline at end of file
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/grid_sample\350\207\252\345\256\232\344\271\211onnx\347\256\227\345\255\220.md" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/grid_sample\350\207\252\345\256\232\344\271\211onnx\347\256\227\345\255\220.md"
new file mode 100644
index 0000000000000000000000000000000000000000..10b653903b8910320f0a0a2a2fc41658769030e5
--- /dev/null
+++ "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/grid_sample\350\207\252\345\256\232\344\271\211onnx\347\256\227\345\255\220.md"
@@ -0,0 +1,7 @@
+[PointRend](https://gitee.com/ascend/modelzoo/pulls/4611)  
+1.onnx不支持grid_sample算子，参考mmcv的自定义算子grid_sample的测试等价代码bilinear_grid_sample进行替换  
+2.去除部分动态shape的where算子  
+3.expand输入类型转换  
+4.div输入类型转换  
+5.padv3算子问题影响网络精度，通过修改代码进行了规避  
+6.本模型是基于detectron2框架的模型  
\ No newline at end of file
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\345\257\274\345\207\272\347\232\204onnx\346\250\241\345\236\213\344\270\215\346\255\243\347\241\256\346\227\266\347\232\204\345\210\206\346\236\220.md" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\345\257\274\345\207\272\347\232\204onnx\346\250\241\345\236\213\344\270\215\346\255\243\347\241\256\346\227\266\347\232\204\345\210\206\346\236\220.md"
new file mode 100644
index 0000000000000000000000000000000000000000..e5c3ed807fcbe07bc47249390d22098da9e6349e
--- /dev/null
+++ "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\345\257\274\345\207\272\347\232\204onnx\346\250\241\345\236\213\344\270\215\346\255\243\347\241\256\346\227\266\347\232\204\345\210\206\346\236\220.md"
@@ -0,0 +1,10 @@
+[refinedet](https://gitee.com/ascend/modelzoo/pulls/4007)  
+1.随机输入导出的onnx使用onnxsim后为空图，模型分段return导出onnx，最终发现随机输入时scores.size(0) == 0导致数据流为空，故导出onnx时使用真实图片作为输入  
+2.nms含有动态shape小算子，替换为自定义batchnms暂未通，因nms在模型最后一部分故可放在后处理  
+3.onnx转om时，不能使用fp16，否则精度不达标  
+  ```
+  --precision_mode allow_fp32_to_fp16
+  ```
+4.常量prior_data在后处理通过代码生成，不用经过模型计算  
+
+[仅供学习](https://gitee.com/wangjiangben_hw/ascend-pytorch-crowdintelligence-doc/tree/master/Ascend-PyTorch%E7%A6%BB%E7%BA%BF%E6%8E%A8%E7%90%86%E6%8C%87%E5%AF%BC/ONNX%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86%E6%8C%87%E5%AF%BC/benchmark/cv/segmentation)
\ No newline at end of file
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\347\233\256\346\240\207\350\267\237\350\270\252SiamMask-\347\275\221\347\273\234\345\210\206\346\256\265.md" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\347\233\256\346\240\207\350\267\237\350\270\252SiamMask-\347\275\221\347\273\234\345\210\206\346\256\265.md"
new file mode 100644
index 0000000000000000000000000000000000000000..89170d2fc9d10b1fcd324cc4976c504c69c100f0
--- /dev/null
+++ "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\347\233\256\346\240\207\350\267\237\350\270\252SiamMask-\347\275\221\347\273\234\345\210\206\346\256\265.md"
@@ -0,0 +1,5 @@
+[SiamMask](https://gitee.com/ascend/modelzoo/pulls/4105)  
+1.om与onnx精度一致。  
+2.因为SiamMask是前后帧连续处理，即上一帧的输入作为下一帧的输出，在Refine模块中进行动态pad，因此模型需要拆分为两段。因为SiamMask部分卷积存在使用自定义kernel来对输入进行卷积操作，导致卷积存在kernel和input的双输入的情况，[issue](http://github.com/onnx/onnx-tensorrt/issues/645)，故在线推理测试性能。计算的是拆分的两部分模型合在一起完整推理的性能。  
+3.因为SiamMask在corr部分固定了reshape之后的形状，并且针对前后帧连续处理，所以模型不支持多batch。  
+4.推理速度较慢的问题目前出在io部分。如果`msame`支持将数据从内存中读写而不必须从`.bin`文件中读写，速度将会进一步加快。  
\ No newline at end of file
diff --git "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\350\266\205\345\210\206\350\276\250\347\216\207\346\250\241\345\236\213\346\241\210\344\276\213.md" "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\350\266\205\345\210\206\350\276\250\347\216\207\346\250\241\345\236\213\346\241\210\344\276\213.md"
index 6836f045378caf74d5dbbfba63542263b779dc44..3a02a098247cbecc5a1f8bd66675519882518bbf 100644
--- "a/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\350\266\205\345\210\206\350\276\250\347\216\207\346\250\241\345\236\213\346\241\210\344\276\213.md"
+++ "b/Ascend-PyTorch\347\246\273\347\272\277\346\216\250\347\220\206\346\214\207\345\257\274/\344\270\223\351\242\230\346\241\210\344\276\213/\345\212\237\350\203\275\346\211\223\351\200\232/\350\266\205\345\210\206\350\276\250\347\216\207\346\250\241\345\236\213\346\241\210\344\276\213.md"
@@ -1 +1,2 @@
-[SRFlow](https://gitee.com/ascend/modelzoo/pulls/3572)
\ No newline at end of file
+[SRFlow](https://gitee.com/ascend/modelzoo/pulls/3572)  
+[Cross-Scale-Non-Local-Attention](https://gitee.com/ascend/modelzoo/pulls/4395)  
\ No newline at end of file