You can get rid of all the training code (because we are focus on 0-shot)
You need to override “def test_step(self, batch, batch_idx):” with our method
Investigate how we can make use of this library for our tests on mvtec's dataset.
Keep in mind we want the ability to evaluate our performances on both non-anomaly and anomaly samples, and separate results by class, possibly even by groups of classes (textures, objects, etc.). How can we load the dataset in a way that will enable this?
How can pre-trained models be used for evaluation?
Gather other useful information
Feel free to comment and document your findings either here or in a separate document.
Duplicate of #7
https://github.com/hcw-00/PatchCore_anomaly_detection
Investigate how we can make use of this library for our tests on mvtec's dataset.
Feel free to comment and document your findings either here or in a separate document.