This code combined two convlstm github codes.
https://github.com/ndrplz/ConvLSTM_pytorch
https://github.com/automan000/Convolutional_LSTM_PyTorch
The code from ndrplz didn't use the ConvlstmCell which is in the original paper https://papers.nips.cc/paper/5955-convolutional-lstm-network-a-machine-learning-approach-for-precipitation-nowcasting.pdf but automan000 did.
So I just change code to make it correctly. I didn't test it on moving mnist dataset. The performance is still unknown. Thanks