Loss function

[likith012]

I am not sure, but I think the loss function is a bit different from the original paper https://github.com/sthalles/PyTorch-BYOL/blob/f007ed6b3619a3ae00eb9c95cde342eac5b98923/trainer.py#L35. In the paper the loss function equation is for unnormalized tensors(loss equation combines both normalization and l2 loss), but here you are normalizing them and applying the loss function.

[Issac-Sun]

I completed a simple implementation on CIFAR10 and the cosine similarity loss in the original article, you can follow https://github.com/Issac-Sun/BYOL_pytorch