0

I would be really thankful if someone can explain to me that how I can build an encoder-decoder model with Tensorflow ConvLSTMCell(), tf.nn.dynamic_rnn(), and tf.contrib.legacy_seq2seq.rnn_decoder(). I would like to build a model that has 3 encoder layers and 3 decoder layers. I have built the model, and I am using Moving mnist dataset as benchmark, in this dataset each sequence has 20 frames, and I am feeding the first 10 frames of each sequence to the encoder and willing to get the next 10 frames as output(prediction), but for the prediction part the model simply tries to output the last input frame (10th frame). In my model I set the number of filters to 128 for the first layers of encoder and decoder and 64, and 64 for the second and third layers respectively. If you want me I can also post the code I wrote here.

MRM
  • 1,099
  • 2
  • 12
  • 29
  • add your code, ask SPECIFIC questions – Mohammad Athar Mar 06 '18 at 02:06
  • Would you please tell me how to make it specific? because I have no error to fix, basically I do not know what is wrong with my model and I posted my code once and there was a comment that it is not minimal. here my other post https://stackoverflow.com/questions/49056451/why-my-convlstm-model-can-not-predict?noredirect=1#comment85117774_49056451 @MohammadAthar – MRM Mar 06 '18 at 02:12

0 Answers0