I have a question. please answer about my question.

[deepvelop99]

While reading and studying this paper, I had a question, which prompted me to reach out. I noticed that the implemented model.py passes the pre-trained WideResNet-50 but lacks information on how to extract Patch Features from the Features. Could you please explain how you extracted the patch features? You mentioned that it passes through WideResNet-50, but I'm curious whether the results are obtained by simply inputting the image into the model or if only the feature extraction part is from a pre-trained model. I'm facing difficulty as there is no explanation provided for (u_i, v_i) in the feature map. I would greatly appreciate your response.

[travishsu]

Hi @deepvelop99,
Actually, I'm not the author of the ReConPatch paper and this repo is not the official one, but the paper mentioned they use the same approach with Patchcore to extract patch features, you can have a look.