Fine-tune inference

[oipa]

Hello,

I have several questions:

1. I see that the function train_and_infer() makes the inference for all the image samples but then saves only the resultant heatmap images for one sample of each anomaly. I guess this happens because of the sampler of the vis_loader. How could I configure the code to save the inference of all the test images?

2. If I introduce a new dataset with an element that only presents one type of anomaly

• The training is done from scratch or is there a transfer learning process?
• Is there any way to try all the different algorithms and decide what works better for this dataset automatically?

Thank you!