gvae.py 文件源码

def _generator(self, z, y, is_training):
        ''' In this version, we only generate the target, so `y` is useless '''
        subnet = self.arch['generator']
        n_layer = len(subnet['output'])
        h, w, c = subnet['hwc']

        # y = tf.nn.embedding_lookup(self.y_emb, y)

        x = self._merge([z, y], subnet['merge_dim'])
        x = lrelu(x)
        with slim.arg_scope(
            [slim.batch_norm],
            scale=True, scope='BN',
            updates_collections=None,
            # decay=0.9, epsilon=1e-5,
            is_training=is_training):

            x = slim.fully_connected(
                x,
                h * w * c,
                normalizer_fn=slim.batch_norm,
                activation_fn=lrelu)

            x = tf.reshape(x, [-1, h, w, c])

            with slim.arg_scope(
                [slim.conv2d_transpose],
                weights_regularizer=slim.l2_regularizer(subnet['l2-reg']),
                normalizer_fn=slim.batch_norm,
                activation_fn=lrelu):

                for i in range(n_layer -1):
                    x = slim.conv2d_transpose(
                        x,
                        subnet['output'][i],
                        subnet['kernel'][i],
                        subnet['stride'][i]
                        # normalizer_fn=None
                        )

                # Don't apply BN for the last layer of G
                x = slim.conv2d_transpose(
                    x,
                    subnet['output'][-1],
                    subnet['kernel'][-1],
                    subnet['stride'][-1],
                    normalizer_fn=None,
                    activation_fn=None)

                # pdb.set_trace()
                logit = x
                # x = tf.nn.tanh(logit)
        # return x, logit
        return tf.nn.sigmoid(logit), logit