CycleGAN.py 文件源码

def deconv2d(input_, output_dim, ks=4, s=2, stddev=0.02, name="deconv2d"):
    with tf.variable_scope(name):
        return slim.conv2d_transpose(input_, output_dim, ks, s, padding='SAME', activation_fn=None,
                                    weights_initializer=tf.truncated_normal_initializer(stddev=stddev),
                                    biases_initializer=None)