Fine-tune Swin

[jedrzejwalega]

Now that we've recreated the authors' ResNet18 results (mostly to confirm the lightning and neptune framework works same way as their skorch) we want to try more recent models than ResNet.

The goal is to fine-tune Swin (pretrained can be taken from timm) on our dataset.

We are benchmarking against the authors' val F1 = 0.91.

[jedrzejwalega]

Swin defined here, but not trying it out for now, since we have ConvNext and ViT to work with.