Models.py

def generator(z,isTrain,keep_prob):
    with tf.variable_scope("generator"):        
        img = tf.layers.dense(z, 6*6*1024, activation = tf.nn.leaky_relu, kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        img = tf.layers.batch_normalization(img, training = isTrain)

        img = tf.reshape(img,[-1,6,6,1024])
        
        img = tf.layers.conv2d_transpose(img, 512, 3, strides = 2, activation = tf.nn.leaky_relu, padding ='same', kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        img = tf.layers.batch_normalization(img, training = isTrain)
        
        img = tf.layers.conv2d_transpose(img, 256, 3, strides = 2, activation = tf.nn.leaky_relu, padding ='same', kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))        img = tf.layers.batch_normalization(img, training = isTrain)
        
        img = tf.layers.conv2d_transpose(img, 128, 3, strides = 2, activation = tf.nn.leaky_relu, padding ='same', kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))        img = tf.layers.batch_normalization(img, training = isTrain)
        
        img = tf.layers.conv2d_transpose(img, 3, 3, strides = 2, padding ='same', name = 'generator_out', kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        img = tf.nn.tanh(img)
        
        return img
        
def discriminator(x,isTrain,keep_prob):
  
    with tf.variable_scope("discriminator"):    
        x = tf.reshape(x,[-1,96,96,3])
        
        logits = tf.layers.conv2d(x,128,5, kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        logits = tf.nn.leaky_relu(logits, alpha = 0.01)
        logits = tf.layers.max_pooling2d(logits,2,2)
        
        logits = tf.layers.conv2d(logits,256,5, kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        logits = tf.nn.leaky_relu(logits, alpha = 0.01)

        logits = tf.layers.max_pooling2d(logits,2,2)
        
        logits = tf.layers.conv2d(logits,512,5, kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        logits = tf.nn.leaky_relu(logits, alpha = 0.01)
        logits = tf.layers.max_pooling2d(logits,2,2)
        
        logits = tf.layers.conv2d(logits,1024,5, kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        logits = tf.nn.leaky_relu(logits, alpha = 0.01)
        logits = tf.layers.max_pooling2d(logits,2,2)
        
        logits = tf.layers.flatten(logits)
        
        logits = tf.layers.dense(logits, 1, name = 'discriminator_out', kernel_initializer=tf.variance_scaling_initializer(dtype = tf.float32))
        
        return logits