solo-learn
solo-learn copied to clipboard
Implement multi-crop (SwAV, DCv2, DINO, MoCo)
and improve SimCLR's implementation
We should keep in mind that this repo has nice parameters for multi-crop on CIFAR
Edit: I have just tested their parameters and they work quite well
@vturrisi @DonkeyShot21 hi guys, I'm trying to use multicrop with DINO but it seems that it does not work. My guess is that it is because timm vision_transformer is used here, but contrary to the "classical" DINO implem, it does not automatically interpolate positional embeddings for different image sizes. Do you relate on your side ?