Patchcore & Fastflow onnx/ov export fails

[pmudgal-Intel]

Describe the bug
Training patchcore or fastflow model fails if export is configured as "Openvino"
python tools/train.py --config anomalib/models/patchcore/config.yaml
In anomalib/models/patchcore/config.yaml file, change export option as "openvino"

Screenshot is attached showing the error. Seems like model is getting trained, but fails to export as onnx.
torch version - 1.11.0+cu102

Screenshots

Hardware and Software Configuration

• OS: [Ubuntu, 20.04]
• NVIDIA Driver Version [470.141.03]
• CUDA Version [e.g. 10.2]
• CUDNN Version [e.g. V10.2.89]
• OpenVINO Version [Optional e.g.v2022.1.0]

[ashwinvaidya17]

@pmudgal-Intel can you have a look at this issue #440? You need to either upgrade torch to 1.12 or replace the cdist operator. Let us know if this works.

[samet-akcay]

@ashwinvaidya17, should we consider replacing cdist by default?

[ashwinvaidya17]

Since this is a recurring question I think it makes sense to replace the function. We can explore the performance difference first.

[pmudgal-Intel]

Patchcore works fine with torch 1.12, but fastflow still shows same error during export.

[samet-akcay]

With the latest anomalib version, PatchCore and FastFlow models can be converted to ONNX and OpenVINO models. I'm now closing this issue. Thanks!