search

HobbitLong / SupContrast

PyTorch implementation of "Supervised Contrastive Learning" (and SimCLR incidentally)
BSD 2-Clause "Simplified" License
3.12k stars 536 forks source link

questions about effect of number of positives #43

Open slchenchn opened 4 years ago

slchenchn commented 4 years ago

Hi, thanks for your nice paper and codes! I have a question after reading the paper, in the section "Effect of Number of Positives", the number of positives always contain one positive which is the same sample with the anchor but different data augmentations, and the remainders are different samples from the same class, have you tried make all the positive from the same sample but different augmentations? Does it improve?

zou-longkun commented 4 years ago

You seem to be talking about self-supervised contrastive learning, right? Just like SimCLR

slchenchn commented 4 years ago

yes, in a self-supervised manner, but with multiple positives which are derived from the same image.