Ashwin Vaidya

@alevangel The thresholds are overwritten when `on_predict_start` is called from `LoadModelCallback`. One idea would be to manually call `torch.load` and pop `LoadModelCallback` from the callbacks list. Then updating the thresholds...

@leibaoer sorry for the late response. As @cytotoxicity8 mentioned, fastflow computes the loss by maximizing the negative log jacobian. You can see it defined here https://github.com/openvinotoolkit/anomalib/blob/main/anomalib/models/fastflow/loss.py#L27. Section 3.3 of the...

@sherpan I can see two projects (padim and dfm) but I cannot see any errors in the output. Also, the reproduce button shows up disabled on my end. 

@sherpan can you confirm if you still see this error?

@sherpan the reproduce button is disabled for me. Maybe because you are the owner of the project. Can you checkout the latest main branch and let us know. I tried...

@jpcbertoldo can you have a look at the issues raised by codacy?

> I'll try if i have time today. Btw tests were failing in my local env. I think it was in the other branch but it wasn't related to my...

- [x] Add benchmarking example on google colab - [x] Add HPO example

Hi @udasinnayan that's a known issue. This might probably be solved when me move towards torchfx but the only option is to use single GPU for now.

I tested this on the current `main` branch and it still does not work with `timm`'s feature extractor. I am closing as it will require a significant change.