Convolutional LSTM in Caffe?

[Ralph Aeschimann]

Given that the RNN and LSTM layers have now been merged to BVLC/caffe master, is there any work being done to add support for a Convolutional LSTM layer?

By Convolutional LSTM, I mean a cell which has convolutional structures in both the input-to-state and state-to-state transitions. 

In comparison, the current LSTM/RNN layers are fully-connected.

See "Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting" by Xingjian Shi as a reference.

[Fatemeh Saleh]

Hi Ralph,

Did you figure out how to use ConvLSTM in Caffe?

[Ram Malav]

Hi All,

I am also looking for a caffe implementation of the ConvLSTM as introduced in paper Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting and  Spatio-temporal video autoencoder with differentiable memory.

I have found this implementation https://github.com/agethen/ConvLSTM-for-Caffe.

Did you find any other implementation? Does this implementation work? Can it be used for many to many mapping?

Please let me know if you have found any other alternative as well.

Thanks.