Actual error message (3.4.17/4.5.5+): DNN/ONNX(Shape): dynamic 'zero' shapes are not supported

• OpenCV => 4.5.1
• Operating System / Platform => Windows 64 Bit
• Compiler => MinGW 7.3

Can opencv load dynamic input models?

An error is reported when using opencv to load the onnx model of dynamic input type.

Detailed description

The following is the compressed package of the input type and model file.

super_resolution.zip

After I converted a pytorch model to onnx, the input size was 1x3x640x640. This model can run at onnxronnuntime. Then I tried to use opencv's readNet(path) function. The model can also work normally.

When I convert this model to dynamic input, this model can also run in onnxronnuntime, but when I use opencv's readNet to load this dynamic input model, ,the program starts to report an error, the error is as follows

OpenCV: terminate handler is called! The last OpenCV error is: OpenCV(4.5.1) Error: Assertion failed (total(os[i]) > 0) in getLayerShapesRecursively, file F:\OpenCV\opencv451\sources\modules\dnn\src\dnn.cpp, line 3520

If opencv can load dynamic input models,do I need to set the input and output size after the model is loaded by opencv?
If yes, is there any information on this part of the content?

Steps to reproduce

I first use python to convert the model to onnx.
The conversion code of pytorch to onnx I used is as follows

 
    input_names = ["input0"]
    output_names = ["output0","output1","output2"]

    x = torch.randn(1, 3, 640, 640, requires_grad=True)
    dynamic_axes = {'input0': {2: "height", 3: 'width'},  # variable lenght axes
                'output0': {1: '_size'},
                'output1': {1: '_size'},
                'output2': {1: '_size'}}
    # Export the model
    torch.onnx.export(net,  # model being run
                  x,  # model input (or a tuple for multiple inputs)
                  "super_resolution.onnx",  # where to save the model (can be a file or file-like object)
                  export_params=True,  # store the trained parameter weights inside the model file
                  opset_version=11,  # the ONNX version to export the model to
                  do_constant_folding=True,  # whether to execute constant folding for optimization
                  input_names=input_names,  # the model's input names
                  output_names=output_names,  # the model's output names
                  dynamic_axes=dynamic_axes) 

Then use opencv to load, the key code is as follows

net = readNet(std::string(pth_file.toLocal8Bit()));

net.setPreferableBackend(DNN_BACKEND_OPENCV);
net.setPreferableTarget(DNN_TARGET_CPU);

cv::Mat blob;
blob = blobFromImage(img,1.0, cv::Size(inpHeight,inpWidth), cv::Scalar(104, 117, 123), true, false, CV_32F);

net.setInput(blob,"input0");

std::vector<cv::Mat> outs;
net.forward(outs, getOutputsNames(net));

std::vector<bbox_l> bbox;

bbox = post_processing(img, outs);

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues,
  answers.opencv.org, Stack Overflow, etc and have not found solution
• I updated to latest OpenCV version and the issue is still there
• There is reproducer code and related data files: videos, images, onnx, etc