farizrahman4u / seq2seq

Sequence to Sequence Learning with Keras
GNU General Public License v2.0
3.17k stars 845 forks source link

Support for ConvLSTM? #196

Open xn8812 opened 7 years ago

xn8812 commented 7 years ago

Hi, I wonder if current package supports ConvLSTM as the recurrent cell? Do you plan to add that? I think this is important for image based applications.

Thanks! @farizrahman4u

xn8812 commented 7 years ago

If I don't want the encoder-decoder structure, I only want to use the previous time-step's output as the current time-step's input (together with some other inputs as well) during the training, do you have any suggestions of how to implement that? The length of the time sequence is a variable.

Can I implement it by using the recurrentshop package? https://github.com/datalogai/recurrentshop

Thanks!