Description

onnx2torch incorrectly omits one weight layer when converting onnx into pytorch. This mis-conversion leads to an IndexError: Dimension out of range in node_converters/global_average_pool.py", line 41, in <lambda>.

Steps to Reproduce

poc.zip

Download and unzip "poc.zip" to get "poc.onnx" file.

Run the following code with poc file.


import onnx, torch
import onnxruntime
from onnx2torch import convert

poc_onnx_model_path = 'poc.onnx'

load onnx_model

onnx_model = onnx.load(poc_onnx_model_path)

check model validity

onnx.checker.check_model(onnx_model)

input

input_torch = torch.randn(1, 3, 512, 512) input_ort = {'input': input_torch.numpy()}

no error in onnx

ort_session = onnxruntime.InferenceSession(poc_onnx_model_path) output_ort = ort_session.run(None, input_ort)

torch_model = convert(onnx_model)

error occurs!

torch_model(input_torch)

Traceback (most recent call last): File "/home/suhwan/grad_course/metadl/report.py", line 23, in torch_model(input_torch) File "/home/suhwan/.local/lib/python3.10/site-packages/torch/fx/graph_module.py", line 738, in call_wrapped return self._wrapped_call(self, *args, kwargs) File "/home/suhwan/.local/lib/python3.10/site-packages/torch/fx/graph_module.py", line 317, in call raise e File "/home/suhwan/.local/lib/python3.10/site-packages/torch/fx/graph_module.py", line 304, in call return super(self.cls, obj).call(*args, *kwargs) # type: ignore[misc] File "/home/suhwan/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(args, kwargs) File "/home/suhwan/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(*args, kwargs) File ".0", line 73, in forward File "/home/suhwan/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(*args, *kwargs) File "/home/suhwan/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(args, kwargs) File "/home/suhwan/.local/lib/python3.10/site-packages/onnx2torch/node_converters/global_average_pool.py", line 46, in forward return forward_lambda() File "/home/suhwan/.local/lib/python3.10/site-packages/onnx2torch/node_converters/global_average_pool.py", line 41, in forward_lambda = lambda: torch.mean(input_tensor, dim=self._x_dims, keepdim=True) IndexError: Dimension out of range (expected to be in range of [-3, 2], but got 3)



### System Configuration
- **onnx2torch version: 1.5.13**
- python version: 3.10.12
- onnx: 1.16.0
- onnxruntime: 1.17.1

ENOT-AutoDL / onnx2torch

onnx2torch incorrectly omits one weight layer when converting onnx into pytorch. #208