Would it be appropriate to apply a checkpoint trained on MVTecAD to our proprietary dataset?

[ZenoVolt]

I was told it's likely possible to run PatchCore or similar models directly on our dataset without training.

However, I haven't been able to get that to work because Pre-trained .ckpt files don't seem to be available — the examples I've seen seem to require training on the dataset first to generate a checkpoint (.ckpt) file, which is then used to run engine.predict().

Some clarity would be helpful, this is probably a dumb question. I am more of a C++ guy than a ML guy so this is new to me.

Thanks!

[abc-125]

Hello, PatchCore should be trained on your data for one epoch to extract features for further comparison with new images. It is not a traditional ML training, but it still requires a training dataset and engine.fit().