Error in infer_shapes.py

[MFJI]

Hello everyone,

I have noticed that I get the following error when running the infer_shapes transform:
File "/workspace/qonnx/src/qonnx/transformation/infer_shapes.py", line 88, in apply model = ModelWrapper(si.infer_shapes(model.model)) File "/opt/conda/lib/python3.8/site-packages/onnx/shape_inference.py", line 41, in infer_shapes inferred_model_str = C.infer_shapes(model_str, check_type, strict_mode, data_prop) onnx.onnx_cpp2py_export.shape_inference.InferenceError: [ShapeInferenceError] (op_type:Transpose, node name: Transpose_7): [ShapeInferenceError] Inferred shape and existing shape differ in dimension 2: (64) vs (56)

Looking at my original ONNX Resnet 18 model I do not understand whether this problem lies with my ONNX model or whether there is a problem with the qonnx import.. For the sake of clarity I have attached my model.

Any help would be very much appreciated.

Kind regards,

Meng Fei
resnet18_img_t.zip

[julienmas]

Hi Meng,

Could you send the onnx file you obtain just before you perform the transformation where you get the error?

[MFJI]

Hello julienmas,

Thank you so much for your reply. Here is the onnx file just before the transformation.
step_tidy_up.zip

[o-ibrahim33]

@julienmas any updates please?

[AliMirhosseini]

I have the same problem with this error
InferenceError: [ShapeInferenceError] (op_type:Transpose, node name: Transpose_11): [ShapeInferenceError] Inferred shape and existing shape differ in dimension 2: (64) vs (32)
when running this line
model = model.transform(Streamline())
did you get the answer?

the model is resnet18 example in brevitas repo and the dataset is CIFAR-10.

[pankajan-T]

Hi, I've got the same issue with my custom resnet model as well. Is there a solution for this issue yet?