def latent(inp, latent_dims=128, name='latent'):
x = tf.keras.layers.Flatten()(inp)
w = tf.keras.layers.Dense(latent_dims, activation='relu')(x)
w_mean = tf.keras.layers.Dense(latent_dims, name='w_mean')(w)
w_log_var = tf.keras.layers.Dense(latent_dims, name='w_log_var')(w)
w = utils.Sampling()([w_mean, w_log_var])
return tf.keras.Model(inputs=inp, outputs=[w, w_mean, w_log_var], name=name)
def latent(inp, latent_dims=128, name='latent', training=True):
x = tf.keras.layers.Conv2D(256, kernel_size=(3, 3), strides=(2, 2), padding='same')(inp)
w = tf.keras.layers.Flatten()(x)
w = tf.keras.layers.Dense(latent_dims)(w)
if training:
w = tf.keras.layers.Dropout(0.4)(w)
w_mean = tf.keras.layers.Dense(latent_dims, name='w_mean')(w)
w_log_var = tf.keras.layers.Dense(latent_dims, name='w_log_var')(w)
w = utils.Sampling()([w_mean, w_log_var])
w = tf.keras.layers.Dropout(0.05)(w)
return tf.keras.Model(inputs=inp, outputs=[w, w_mean, w_log_var], name=name)