I would like to have a small example of building an encoder-decoder network using Tensorflow ConvLSTMCell. Thanks
Asked
Active
Viewed 544 times
2
-
Why isn't replacing GRU with ConvLSTM working? – Maxim Mar 08 '18 at 19:41
-
I don't get your question @Maxim. I want to use convlstm – MRM Mar 09 '18 at 19:48
-
I mean there are lots of examples using LSTM and GRU. Have you tried to simply replace the cell in them? – Maxim Mar 09 '18 at 22:39
-
@Maxim Would you please put a link to one of those examples here. I actually built a model, but the thing is with ConvLSTMCell when we have a different number of filters in each layer, the output of one layer to the next layer does not match (When I use a loop_function) in the decoder. Basically, I want to know how should I make these connections so I build a model which can predict. – MRM Mar 10 '18 at 05:20
-
Check this repo - https://github.com/maxim5/tensorflow-seq2seq-translate It's encoder-decoder network for machine translation, uses GRU cells, adapted to latest tensorflow. The main model is in `seq2seq_model.py` – Maxim Mar 10 '18 at 07:28
-
It's convolutional, there could be differences. How does / does pooling happen between convLSTM cells in a multi layer network? Are there strides? – dasWesen Jul 16 '18 at 16:28