manugarri / keras_monet Goto Github PK

Is there a trick to changing the Quantizing of the input images to accommodate for larger pixel size images while not increasing the RAM usage dramatically?

Theano and keras are installed from their git repos. When running "python lstm_image_generation.py", the script stopped at this line:
model.add(LSTM(len(chars), 512, return_sequences=True))
I believe that because the API change in Keras.

Hi, I am a newbee into rnn. As you see, theano sometimes not easy for configuration. My gpu is gtx760 and it use nearly 800s per iteration.
I am afraid this is too slow, so I wanan know which gpu you used for this code(used nearly 350s per iteration) and check if I make something not appropriate for the theano configuration.