ConvLSTM3D in pytorch for 3D data

Hi is there anything in pytorch like class ConvLSTM2D(ConvRNN2D): as in keras.
I want a convLSTM for 3D data which is still not supported in keras, I could find ConvLSTM2D which is meant for 2D data,
Please suggest me if there is ConvLSTM3D already supported in pytorch , if not how to implement it.

The easiest way is to spell out the LSTM cell yourself.
PyTorch lets you make this fast, look at the custom LSTMs we use in benchmarking for inspiration.

Best regards


1 Like

Thanks tom for your kind reply, I had a look at custom LSTM cell , but I need some lstm cell which works as convLstm, If I take custom LSTM as a basic block it will require a lot of work around to convert it to convLstm and convLstm3D seems to be a long way to go.
Any ways thanks again I will start the implementation of the same

While your question is from last year, I still wanted to chime in to maybe provide some help.

If you check the answer at Passing hidden layers to ConvLSTM, I think you should be able to modify the code to utilize Conv3d layers instead of their 2D counterparts.