thomas-beznik opened a new issue #17848: Import ONNX model from Pytorch to MXNet URL: https://github.com/apache/incubator-mxnet/issues/17848 ## Description Hello, I am trying to convert an ONNX model coming from a PyTorch model to an MXNet model. The PyTorch model is an FPN that comes from this repository https://github.com/qubvel/segmentation_models.pytorch. I am running into the following error when using the method mxnet.contrib.onnx.onnx2mx.import_model ### Error Message Traceback (most recent call last): File "C:/Users/thoma/Documents/relu/AI_API/API/Detectors/Slice2DDetectorONNX.py", line 96, in <module> sym, arg_params, aux_params = import_model(model_path) File "C:\Users\thoma\Documents\relu\AI_API\env\lib\site-packages\mxnet\contrib\onnx\onnx2mx\import_model.py", line 59, in import_model sym, arg_params, aux_params = graph.from_onnx(model_proto.graph) File "C:\Users\thoma\Documents\relu\AI_API\env\lib\site-packages\mxnet\contrib\onnx\onnx2mx\import_onnx.py", line 115, in from_onnx inputs = [self._nodes[i] for i in node.input] File "C:\Users\thoma\Documents\relu\AI_API\env\lib\site-packages\mxnet\contrib\onnx\onnx2mx\import_onnx.py", line 115, in <listcomp> inputs = [self._nodes[i] for i in node.input] KeyError: 'components.0.net.net.encoders.0.basic_module.SingleConv1.conv.weight' I saw similar issues (such as https://github.com/apache/incubator-mxnet/issues/13395), which say that the problem comes from the fact that Dynamic Shape isn't implemented. But in my case, I am not using a dynamic shape; all my inputs are resized to a fixed shape before being fed to the model. Is there thus a way to easily change my architecture such that it uses this fixed shape and enables me to load my model in MXNet? I can give more information if needed, any direction to solve this would be greatly appreciated! ## To Reproduce - python 3.6 - onnx 1.6.0 - mxnet 1.5.0
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
