lucidrains/byol-pytorch

For BYOL, cosine similarity of two images should be evaluated by online_encoder or target_encoder?

vincezengqiang opened this issue · 1 comments

vincezengqiang commented

cosine similarity of two images should be evaluated by online_encoder or target_encoder?
if should be evaluated by online_encoder, should be evaluated before online_predictor or after online_predictor?

vincezengqiang commented

i find the answer in paper,