# training data sources config.image_dir = os.path.join(os.pardir, 'data', FLAGS.image_dir) config.image_ext = '*.png' config.img_verbose = True # model configurations config.batch_size = FLAGS.batch_size config.z_dim = 100 # number inputs to gener config.c_dim = 1 config.gf_dim = FLAGS.gf_dim # number of gener conv filters config.df_dim = FLAGS.df_dim # number of discim conv filters config.gfc_dim = 1024 # number of gener fully connecter layer units config.dfc_dim = 1024 # number of discim fully connected layer units config.alpha = 0.1 # leaky relu alpha config.batch_norm = True config.minibatch_discrim = True # training hyperparameters config.epoch = FLAGS.epoch config.learning_rate = FLAGS.learning_rate # optim learn rate config.beta1 = FLAGS.beta1 # momentum config.repeat_data = True config.shuffle_data = True config.buffer_size = 4 config.drop_remainder = True # currently fails if false! config.gener_iter = FLAGS.gener_iter # times to update generator per discriminator update config.noisy_inputs = False # add some small noise to the input images config.flip_inputs = False # whether to flip the black white pixels # i/o structures