CheXNet fine tunning #5
Description
The zero-shot task is super funny, but in practice, transfer learning is usually done. It would be interesting to see how good medclip is relative to other backbones.
At the moment the entire pipeline can be carried out using the original CLIP weights and as soon as we have the finished model, we can change the weights and perform the same evaluation.