I tried to convert a torch RNN net into pytorch. https://github.com/Andybert/Recurrent-Convolutional-Video-ReID
The pytorch net works and converges, but it have a performance loss(rank 1: 56 vs 40). I don’t know why.
any advices? or someone can help me check whether the model defined in pytorch is consistent with the model defined in torch, especially the rnn module？