suminRoh

Hi, I have a question. Doesn't (k+1) appear in mask_pos_view.shape[0] in the code? I think the correct code is: loss_target = - torch.sum((mask_pos_view_target * log_prob).sum(1) / mask_pos_view.sum(1))

Why do I have to divide by the batch size? In main_moco_supcon_imaba.py, does not the AverageMeter of losses compute the average loss? Also, if I have to divide by the...

I understand. Thank you for explaining in detail !