zsnoob

My idea is just like yours. After debugging, I found that during the training epoch, all GPUs compute the same global loss with the same sim_matrix instead of individually calculating...

> As far as I understand: > > * Pair: number of samples (videos) > * L: length of video in frames > * T: time axis? It's always 1...

Also mentioned in this issue, https://github.com/ArrowLuo/CLIP4Clip/issues/101#issue-1607949547.