DCCC-LSTM

cs570_2017_spring_term project

Required

mnist.h5 (need to be installed & located at "data/")

Folders

DC-LSTM/ Deep ConvLSTM

DCCC-LSTM/ Deep Conditioned Composite ConvLSTM

DCComp-LSTM/ Deep Composite ConvLSTM

DCCond-LSTM/ Deep Conditioned ConvLSTM

GAN/ GAN

basicLSTM/ simple single layer LSTM

Weight save & load

How to save

• train_example: will be saved into weights folder
• train_example_conv: will be save into weights_conv folder

How to restore weights

argv[1] gets a weight file path, without a file extension.

loss~.~~~~~~

$ python train_example.py weights/2017-05-10_22:51:26.218343__step0__loss2839.130859
$ python train_example_conv.py weights_conv/2017-05-10_22:51:26.218343__step0__loss2839.130859

Video Evaluation

$ python evaluation.py [weight-set]
$ python evaluation_conv.py [weight-set]

Video folder will be created on execution. Videos will be saved into the folder.

Test data set : https://drive.google.com/open?id=0B4DyTsz0E9HFakRhb2FPVzJieUE We edited the original moving mnist dataset so that the sequency of array is different for our convenience.