[Bug]: torch.onnx.export fails for higher versions of torch

[ORippler]

Describe the bug

Apparently torch performs runtime type-checking with beartype now (not sure since what version exactly). As a consequence, torch.onnx.export errors during model export since we pass a Path and not str | io.BytesIO types to it.

File <@beartype(torch.onnx.utils.export) at 0x7f496fa14f70>:82, in export(__beartype_func, __beartype_get_violation, __beartype_object_139953382171520, __beartype_object_94218987035792, __beartype_object_139953391361152, __beartype_object_94218985712112, __beartype_object_139957334026176, __beartype_object_94218917833984, __beartype_getrandbits, __beartype_object_94218985711152, __beartype_object_139953396621568, __beartype_object_94218917827408, __beartype_object_139953387959104, __beartype_object_94218918055504, *args, **kwargs)

BeartypeCallHintParamViolation: @beartyped torch.onnx.utils.export() parameter f="PosixPath('/workspaces/ml-docker-dev/anomalib/asdf/results/patchcore/terminal bloc... violates type hint typing.Union[str, _io.BytesIO], as <protocol "pathlib.PosixPath"> "PosixPath('/workspaces/ml-docker-dev/anomalib/asdf/results/patchcore/terminal bloc... not str or <class "_io.BytesIO">.

Can be easily fixed by casting to str in torch.onnx.export

Dataset

N/A

Model

N/A

Steps to reproduce the behavior

1. install anomalib from master
2. try to export to onnx

OS information

OS information:

• OS: [e.g. Ubuntu 20.04]
• Python version: [e.g. 3.8.10]
• Anomalib version: [e.g. 0.3.6]
• PyTorch version: [e.g. 1.9.0]
• CUDA/cuDNN version: [e.g. 11.1]
• GPU models and configuration: [e.g. 2x GeForce RTX 3090]
• Any other relevant information: [e.g. I'm using a custom dataset]

Expected behavior

Export works

Screenshots

No response

Pip/GitHub

pip

What version/branch did you use?

No response

Configuration YAML

n/a

Logs

File <@beartype(torch.onnx.utils.export) at 0x7f496fa14f70>:82, in export(__beartype_func, __beartype_get_violation, __beartype_object_139953382171520, __beartype_object_94218987035792, __beartype_object_139953391361152, __beartype_object_94218985712112, __beartype_object_139957334026176, __beartype_object_94218917833984, __beartype_getrandbits, __beartype_object_94218985711152, __beartype_object_139953396621568, __beartype_object_94218917827408, __beartype_object_139953387959104, __beartype_object_94218918055504, *args, **kwargs)

BeartypeCallHintParamViolation: @beartyped torch.onnx.utils.export() parameter f="PosixPath('/workspaces/ml-docker-dev/anomalib/asdf/results/patchcore/terminal bloc... violates type hint typing.Union[str, _io.BytesIO], as <protocol "pathlib.PosixPath"> "PosixPath('/workspaces/ml-docker-dev/anomalib/asdf/results/patchcore/terminal bloc... not str or <class "_io.BytesIO">.

### Code of Conduct

- [X] I agree to follow this project's Code of Conduct

[ORippler]

No clue why mypy doesn't detect this though, and the type-error is caught only at run-time. Might be worth digging deeper into this.

[samet-akcay]

that's odd, mypy should have caught in the first place. Thanks for spotting this

[Kyle-fang]

Hello! I'm using onnx obtained from anomalib and then reasoning with onnxruntime, the input image shape is [1, 3, 256, 256] but the output is [1, 1, 256, 256], isn't it supposed to be the number of categories? Is there a problem with the .onnx file? @samet-akcay @ORippler @Lothiraldan

[ORippler]

Hello! I'm using onnx obtained from anomalib and then reasoning with onnxruntime, the input image shape is [1, 3, 256, 256] but the output is [1, 1, 256, 256], isn't it supposed to be the number of categories? Is there a problem with the .onnx file? @samet-akcay @ORippler @Lothiraldan

The output is a heatmap, so everything is working as expected