some LSTM network problemse

[梅骁]

Hi,

     I'm a c++ programmer and I really want to use caffe. I noticed the implementation of LSTM layer is constraint by constant time step, and I don't know why to do that. If I'm going to analysis videos, do I have to sample frames?

     I want to implement a LSTM with arbitrary time step in caffe. Can you give me some tips? It seems need to take some efforts for template models like this.

     Another question, if I want to implement a encoder-decoder for video analysis, how should I write the configure file.

     Thank you. Looking forward to your reply!