ContrastiveLoss problem

[zhangjiwei-japan]

I think in the code “nominator = torch.exp(positives / self.temperature) # 2bs denominator = self.negatives_mask torch.exp(similarity_matrix / self.temperature),” denominator contains nominator and the formula (k!=i) in the paper is somewhat different，thank you.

[Sara-Ahmed]

Sorry for the late response. The loss function is updated in the main file and the Arxiv paper will be updated soon.