Error(s) in loading state_dict for PatchcoreLightning

Describe the bug

Running tools/inference/torch_inference.py throws:

Traceback (most recent call last):
  File "/workspace/tools/inference/torch_inference.py", line 97, in <module>
    infer()
  File "/workspace/tools/inference/torch_inference.py", line 73, in infer
    inferencer = TorchInferencer(config=args.config, model_source=args.weights)
  File "/workspace/anomalib/deploy/inferencers/torch_inferencer.py", line 54, in __init__
    self.model = self.load_model(model_source)
  File "/workspace/anomalib/deploy/inferencers/torch_inferencer.py", line 85, in load_model
    model.load_state_dict(torch.load(path)["state_dict"])
  File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1497, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for PatchcoreLightning:
	Unexpected key(s) in state_dict: "normalization_metrics.min", "normalization_metrics.max".

To Reproduce

1. Using default patchcore/config.yaml
2. Train the model: python tools/train.py --config anomalib/models/patchcore/config.yaml
3. Use trained model for torch_inference:

python tools/inference/torch_inference.py \
--config anomalib/models/patchcore/config.yaml \
--weights results/patchcore/mvtec/bottle/weights/model-v14.ckpt \
--input datasets/MVTec/bottle/test/broken_large/000.png \
--output results/patchcore/mvtec/bottle/images

Additional context

Error is present since commit d6951ebaaf36477bf245b17b5b6a563d35892e81

[alevangel]

@alexriedel1 oh gosh! It was a messy with the colab versions, I found the problem where was… In the big block of isntalling libraries I also installed anomalib (while also cloning the git repo), and this created an overlapping in the versions.

Now it works 😂 Thanks!!

[alexriedel1]

On loading the state dictionary, it looks like the saved model file was trained before using timm models for feature extraction: Unexpected key(s) in state_dict: "model.feature_extractor.backbone.layer1.0.bn2.running_var" https://github.com/openvinotoolkit/anomalib/blob/e4f849944c67743491d5aa46645c671737b51e1c/anomalib/models/components/feature_extractors/feature_extractor.py#L39

Where as the model defined for loading the weights is the most recent one using the timm feature extractor: Missing key(s) in state_dict: "model.feature_extractor.feature_extractor.layer1.0.bn2.running_var" https://github.com/openvinotoolkit/anomalib/blob/bd369190cdb49f22a22ebd058ea4af46f50aee26/anomalib/models/components/feature_extractors/feature_extractor.py#L43-L49

Are you sure you trained the model with the most recent version of anomalib?