Closed fanrz closed 2 years ago
Sorry, we arrange the code after the paper being accepted. There may be the gap between the paper and code. You can change the code to "positive = torch.mean(score[:, :, :15], dim=2) negative = torch.mean(score[:, :, 50:65], dim=2) # [N,H*W] ".
thank for your knid reply. I want to know for what reason, you pick 15 50 65. any explanation for the values? or observation of the training?
observation of the training
great, thanks
thanks for your work. I have on question. in your paper, For each ordered sequence, we take the top n1N related ones as relevant features and n1N unrelated ones starting from n2N as irrelevant features. u use
positive = torch.mean(score[:, :, :5], dim=2) negative = torch.mean(score[:, :, 50:], dim=2) # [N,H*W]
why do you use 5 and 50?