I have problem of loading onnx model by opencv. How to fix it? Thx.
pytorch 1.7.1
opencv 4.3.0
Here is the detail:

python main_yolov5.py --net_type yolov5s
Traceback (most recent call last):
File "main_yolov5.py", line 110, in
yolonet = yolov5(args.net_type, confThreshold=args.confThreshold, nmsThreshold=args.nmsThreshold, objThreshold=args.objThreshold)
File "main_yolov5.py", line 21, in init
self.net = cv2.dnn.readNet(yolo_type + '.onnx')

cv2.error: OpenCV(4.3.0) /io/opencv/modules/dnn/src/graph_simplifier.cpp:79: error: (-212:Parsing error) Input node with name 925 not found in function 'getInputNodeId'

I train my model with your changes, y test this using onnx runtime and all is ok , but in OpenCV I get this error:
outs[row_ind:row_ind+length, 0:2] = (outs[row_ind:row_ind+length, 0:2] * 2. - 0.5 + np.tile(self.grid[i],(self.na, 1))) * int(self.stride[i])
ValueError: operands could not be broadcast together with shapes (1,2,8) (19200,2)

你好，你的代码跑通了，模型是yolov5哪个版本。
ultralytics/yolov5-5.0的模型能跑吗，需要改代码吗

试了一下效果挺好的。请问如果把Sigmoid那层放入onnx来做，是不是更好些？

请问一下有推荐的硬件来部署吗 感谢！！

TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!

相关代码：route = F.interpolate(xRt2, size=(int(xRt2.shape[2] * 2), int(xRt2.shape[3] * 2)), mode='nearest')

大佬 请问这个警告影响推理吗？

Excuse me！How to train my own dataset and convert to onxx ？

Where is yolov5s_param.pth ?

I cannot understand the formulation：
outs[row_ind:row_ind+length, 0:2] = (outs[row_ind:row_ind+length, 0:2] * 2. - 0.5 + np.tile(self.grid[i],(self.na, 1))) * int(self.stride[i])
outs[row_ind:row_ind+length, 2:4] = (outs[row_ind:row_ind+length, 2:4] * 2) ** 2 * np.repeat(self.anchor_grid[i],h*w, axis=0)
would you please help explain it

您导出的yolov5s.onnx的推理结果和原版python差别很大
会是什么原因造成的呢
谢谢

您好，大佬：
在这个 fork 中，我根据运行您另一个仓库https://github.com/hpc203/yolov5-face-landmarks-opencv-v2 里的main_export_onnx.py文件调整了一些参数以及更新了common.py部分代码，成功转换出yolov5s.onnx并读取成功，实际测试时在 1660 Ti max q 上跑纯推理平均10ms，同时运行浏览器看视频的话12ms。

然而，在转换yolov5n的时候出现了如下错误：

RuntimeError: Given groups=1, weight of size [32, 128, 1, 1], expected input[1, 256, 26, 26] to have 128 channels, but got 256 channels instead

部分Trace:
...File "F:\Downloads\yolov5-dnn-cpp-python-v2-main\convert-onnx\common.py", line 41, in forward
return self.act(self.bn(self.conv(x)))
...File "F:\Downloads\yolov5-dnn-cpp-python-v2-main\convert-onnx\convert_onnx.py", line 107, in
torch.onnx.export(onnx_model, inputs, output_onnx, verbose=False, opset_version=12, input_names=['images'], output_names=['out'])
...File "F:\Downloads\yolov5-dnn-cpp-python-v2-main\convert-onnx\yolov5n.py", line 38, in forward
x = self.seq13_C3(x)
...File "F:\Downloads\yolov5-dnn-cpp-python-v2-main\convert-onnx\common.py", line 248, in forward
return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), dim=1))

其他错误都是torch包里的。

以及一些运行时print的：

Namespace(net_type='yolov5n', num_classes=2)
349 348
13.cv1.conv.weight backbone_head.seq13_C3.cv1.conv.weight torch.Size([64, 256, 1, 1]) torch.Size([32, 128, 1, 1])
13.cv1.bn.weight backbone_head.seq13_C3.cv1.bn.weight torch.Size([64]) torch.Size([32])
13.cv1.bn.bias backbone_head.seq13_C3.cv1.bn.bias torch.Size([64]) torch.Size([32])
13.cv1.bn.running_mean backbone_head.seq13_C3.cv1.bn.running_mean torch.Size([64]) torch.Size([32])
13.cv1.bn.running_var backbone_head.seq13_C3.cv1.bn.running_var torch.Size([64]) torch.Size([32])
13.cv2.conv.weight backbone_head.seq13_C3.cv2.conv.weight torch.Size([64, 256, 1, 1]) torch.Size([32, 128, 1, 1])
13.cv2.bn.weight backbone_head.seq13_C3.cv2.bn.weight torch.Size([64]) torch.Size([32])
13.cv2.bn.bias backbone_head.seq13_C3.cv2.bn.bias torch.Size([64]) torch.Size([32])
13.cv2.bn.running_mean backbone_head.seq13_C3.cv2.bn.running_mean torch.Size([64]) torch.Size([32])
13.cv2.bn.running_var backbone_head.seq13_C3.cv2.bn.running_var torch.Size([64]) torch.Size([32])
13.cv3.conv.weight backbone_head.seq13_C3.cv3.conv.weight torch.Size([128, 128, 1, 1]) torch.Size([64, 64, 1, 1])
13.cv3.bn.weight backbone_head.seq13_C3.cv3.bn.weight torch.Size([128]) torch.Size([64])
13.cv3.bn.bias backbone_head.seq13_C3.cv3.bn.bias torch.Size([128]) torch.Size([64])
13.cv3.bn.running_mean backbone_head.seq13_C3.cv3.bn.running_mean torch.Size([128]) torch.Size([64])
13.cv3.bn.running_var backbone_head.seq13_C3.cv3.bn.running_var torch.Size([128]) torch.Size([64])
13.m.0.cv1.conv.weight backbone_head.seq13_C3.m.0.cv1.conv.weight torch.Size([64, 64, 1, 1]) torch.Size([32, 32, 1, 1])
13.m.0.cv1.bn.weight backbone_head.seq13_C3.m.0.cv1.bn.weight torch.Size([64]) torch.Size([32])
13.m.0.cv1.bn.bias backbone_head.seq13_C3.m.0.cv1.bn.bias torch.Size([64]) torch.Size([32])
13.m.0.cv1.bn.running_mean backbone_head.seq13_C3.m.0.cv1.bn.running_mean torch.Size([64]) torch.Size([32])
13.m.0.cv1.bn.running_var backbone_head.seq13_C3.m.0.cv1.bn.running_var torch.Size([64]) torch.Size([32])
13.m.0.cv2.conv.weight backbone_head.seq13_C3.m.0.cv2.conv.weight torch.Size([64, 64, 3, 3]) torch.Size([32, 32, 3, 3])
13.m.0.cv2.bn.weight backbone_head.seq13_C3.m.0.cv2.bn.weight torch.Size([64]) torch.Size([32])
13.m.0.cv2.bn.bias backbone_head.seq13_C3.m.0.cv2.bn.bias torch.Size([64]) torch.Size([32])
13.m.0.cv2.bn.running_mean backbone_head.seq13_C3.m.0.cv2.bn.running_mean torch.Size([64]) torch.Size([32])
13.m.0.cv2.bn.running_var backbone_head.seq13_C3.m.0.cv2.bn.running_var torch.Size([64]) torch.Size([32])
anchor
24.m.0.weight yolo_layers.m.0.bias torch.Size([255, 64, 1, 1]) torch.Size([21])
24.m.0.bias yolo_layers.m.1.weight torch.Size([255]) torch.Size([21, 128, 1, 1])
24.m.1.weight yolo_layers.m.1.bias torch.Size([255, 128, 1, 1]) torch.Size([21])
24.m.1.bias yolo_layers.m.2.weight torch.Size([255]) torch.Size([21, 256, 1, 1])
24.m.2.weight yolo_layers.m.2.bias torch.Size([255, 256, 1, 1]) torch.Size([21])
349 348
13.cv1.conv.weight backbone.seq13_C3.cv1.conv.weight torch.Size([64, 256, 1, 1]) torch.Size([32, 128, 1, 1])
13.cv1.bn.weight backbone.seq13_C3.cv1.bn.weight torch.Size([64]) torch.Size([32])
13.cv1.bn.bias backbone.seq13_C3.cv1.bn.bias torch.Size([64]) torch.Size([32])
13.cv1.bn.running_mean backbone.seq13_C3.cv1.bn.running_mean torch.Size([64]) torch.Size([32])
13.cv1.bn.running_var backbone.seq13_C3.cv1.bn.running_var torch.Size([64]) torch.Size([32])
13.cv2.conv.weight backbone.seq13_C3.cv2.conv.weight torch.Size([64, 256, 1, 1]) torch.Size([32, 128, 1, 1])
13.cv2.bn.weight backbone.seq13_C3.cv2.bn.weight torch.Size([64]) torch.Size([32])
13.cv2.bn.bias backbone.seq13_C3.cv2.bn.bias torch.Size([64]) torch.Size([32])
13.cv2.bn.running_mean backbone.seq13_C3.cv2.bn.running_mean torch.Size([64]) torch.Size([32])
13.cv2.bn.running_var backbone.seq13_C3.cv2.bn.running_var torch.Size([64]) torch.Size([32])
13.cv3.conv.weight backbone.seq13_C3.cv3.conv.weight torch.Size([128, 128, 1, 1]) torch.Size([64, 64, 1, 1])
13.cv3.bn.weight backbone.seq13_C3.cv3.bn.weight torch.Size([128]) torch.Size([64])
13.cv3.bn.bias backbone.seq13_C3.cv3.bn.bias torch.Size([128]) torch.Size([64])
13.cv3.bn.running_mean backbone.seq13_C3.cv3.bn.running_mean torch.Size([128]) torch.Size([64])
13.cv3.bn.running_var backbone.seq13_C3.cv3.bn.running_var torch.Size([128]) torch.Size([64])
13.m.0.cv1.conv.weight backbone.seq13_C3.m.0.cv1.conv.weight torch.Size([64, 64, 1, 1]) torch.Size([32, 32, 1, 1])
13.m.0.cv1.bn.weight backbone.seq13_C3.m.0.cv1.bn.weight torch.Size([64]) torch.Size([32])
13.m.0.cv1.bn.bias backbone.seq13_C3.m.0.cv1.bn.bias torch.Size([64]) torch.Size([32])
13.m.0.cv1.bn.running_mean backbone.seq13_C3.m.0.cv1.bn.running_mean torch.Size([64]) torch.Size([32])
13.m.0.cv1.bn.running_var backbone.seq13_C3.m.0.cv1.bn.running_var torch.Size([64]) torch.Size([32])
13.m.0.cv2.conv.weight backbone.seq13_C3.m.0.cv2.conv.weight torch.Size([64, 64, 3, 3]) torch.Size([32, 32, 3, 3])
13.m.0.cv2.bn.weight backbone.seq13_C3.m.0.cv2.bn.weight torch.Size([64]) torch.Size([32])
13.m.0.cv2.bn.bias backbone.seq13_C3.m.0.cv2.bn.bias torch.Size([64]) torch.Size([32])
13.m.0.cv2.bn.running_mean backbone.seq13_C3.m.0.cv2.bn.running_mean torch.Size([64]) torch.Size([32])
13.m.0.cv2.bn.running_var backbone.seq13_C3.m.0.cv2.bn.running_var torch.Size([64]) torch.Size([32])
anchor
24.m.0.weight m0.bias torch.Size([255, 64, 1, 1]) torch.Size([21])
24.m.0.bias m1.weight torch.Size([255]) torch.Size([21, 128, 1, 1])
24.m.1.weight m1.bias torch.Size([255, 128, 1, 1]) torch.Size([21])
24.m.1.bias m2.weight torch.Size([255]) torch.Size([21, 256, 1, 1])
24.m.2.weight m2.bias torch.Size([255, 256, 1, 1]) torch.Size([21])

请问我应该如何调整代码或网络参数来避免这个错误呢？

Hi!

I'm getting the following error:

$ python3 main_yolov5.py --net_type yolov5n
Traceback (most recent call last):
  File "main_yolov5.py", line 111, in <module>
    dets = yolonet.detect(srcimg)
  File "main_yolov5.py", line 95, in detect
    outs[row_ind:row_ind+length, 0:2] = (outs[row_ind:row_ind+length, 0:2] * 2. - 0.5 + np.tile(self.grid[i],(self.na, 1))) * int(self.stride[i])
ValueError: operands could not be broadcast together with shapes (1,2,85) (19200,2) 

I'm using OpenCV 4.5.3 and Python 3.8.10

My yolov5s.onnx file was generated using:

git clone https://github.com/ultralytics/yolov5
cd yolov5
pip install -r requirements.txt
python3 export.py --weights yolov5s.pt --img 640 --include onnx

Any ideas what could be wrong?

/media/yzl/C4A8E8FBE4769FDE/sr/cpp/opencv-4.2.0/modules/dnn/src/onnx/onnx_importer.cpp:327: error: (-215:Assertion failed) model_proto.has_graph() in function 'populateNet'