Retrain PatchCore

[Rphilips05]

Hi, I have a question.

If you have already trained a PatchCore model and it returns an anomaly while a human expert knows it is not an anomaly. How would you give the model feedback so it won't make the same mistake again.

If I am correct you need to retrain the whole model and can't just add the one image. I could add it to the old dataset en retrain the model but i'll doubt that it have the right impact. Could I add more weight to this one image through augmentations or other options? And if so is this beneficial for the model or is this not recommended? The end goal is that a human could give feedback if the model is wrong.