def generator(flow):
flow = fullyConnected('layer_0', flow, 1024, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = fullyConnected('layer_1', flow, 7 * 7 * 64, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = tf.reshape(flow, [batch_size, 7, 7, 64])
flow = deconv('layer_2', flow, [batch_size, 14, 14, 32], 5, 2)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = tf.nn.sigmoid(
deconv('layer_3', flow, [batch_size, 28, 28, 1], 5, 2))
return flow
def discriminator(flow):
flow = conv('layer_0', flow, 32, 5, 2, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = conv('layer_1', flow, 64, 5, 2, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = flat(flow)
flow = fullyConnected('layer_2', flow, 1024, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
# flow = tf.nn.dropout(flow, .5)
flow = fullyConnected('output', flow, 1, None)
return flow
def encoder(flow):
flow = conv('layer_0', flow, 32, 5, 2, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = conv('layer_1', flow, 64, 5, 2, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
flow = flat(flow)
flow = fullyConnected('layer_2', flow, 1024, None)
flow = normalizeBatch(flow, True)
flow = lrelu(flow)
mean = fullyConnected('mu', flow, latent_dim, None)
sigma = fullyConnected('sigma', flow, latent_dim, None)
return mean, sigma