Data Augumentation not same with BYOL?

lucidrains / pixel-level-contrastive-learning

Implementation of Pixel-level Contrastive Learning, proposed in the paper "Propagate Yourself", in Pytorch

MIT License

252 stars 29 forks source link

Closed guanfuchen closed 3 years ago

guanfuchen commented 3 years ago

color jitter and gaussian kernel, sigma is not same with BYOL, does the paper use the special parameter?

lucidrains commented 3 years ago

@guanfuchen i'm actually using the set from SimCLR i believe, but let me check on this later and get back to you!

lucidrains commented 3 years ago

@guanfuchen just like my other libraries, you can customize your own augmentation functions by passing in augment_fn and augment_fn2

guanfuchen commented 3 years ago

I refer the other implementaion. Thxs. @lucidrains