a issue about negative samples in logits

[haofuly]

Thanks for your great work!

social-nce/crowd_nav/snce/contrastive.py

Line 68 in 6163060

logits = torch.cat([sim_pos.view(-1).unsqueeze(1), sim_neg.view(sim_neg.size(0), -1).repeat_interleave(self.horizon, dim=0)], dim=1) / self.temperature

I am not sure why "sim_neg.view(sim_neg.size(0), -1).repeat_interleave(self.horizon, dim=0)" rather than "sim_neg.view(sim_neg.size(0) * sim_neg.size(1), -1)". The number of negative samples corresponding to one positive sample should be (N-1)*8 or H*(N-1)*8?