So I have 3 PyTorch models where 2 of them give me trouble when loading in ONNX, the inference works fine in PyTorch and the conversion is all ok too.
So in the first model I get this error when reading the ONNX model in OpenCV:
Assertion failed (inputs.size()) in cv::dnn::dnn4_v20190122::Layer::getMemoryShapese
With the other model I get this error:
Blob 64 not found in const blobs
This has all something to do with the input where in the first error the inputs.size() is 0 and in the second model the blob isn’t big enough somehow.
I just don’t know where the error lies, as the PyTorch model alone is valid, I suppose it’s either in the PyTorch → ONNX model conversion or in OpenCV itself.
Here is the project including the first onnx model file GitHub - Aeroxander/decodererror