生成對抗網路GAN目前在圖片生成以及對抗訓練上取得了非常好的應用，本文旨在做一個簡單的tf wgan-gp mnist 生成教程，所使用的程式碼非常簡單，希望和大家共同學習。程式碼如下：
所使用的環境：
tensorflow 1.2.0
GPU加速，CPU上也是可以的，就是很慢，可以把batchsize改小，用cpu好訓練一些，順便把生成影象程式碼處改一下，我的batchsize64，save_images的引數是[8，8]，如果batchsize=16，就改為[4，4]

#coding:utf-8
import os
import numpy as np
import scipy.misc
import
 
 tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data #as mnist_data

def conv2d(name, tensor,ksize, out_dim, stddev=0.01, stride=2, padding='SAME'):
    with tf.variable_scope(name):
        w = tf.get_variable('w', [ksize, ksize, tensor.get_shape()[-1],out_dim], dtype=tf.float32,
                            initializer=tf.random_normal_initializer(stddev=stddev))
        var = tf.nn.conv2d(tensor,w,[1
 
,stride, stride,1],padding=padding)
        b = tf.get_variable('b', [out_dim], 'float32',initializer=tf.constant_initializer(0.01))
        return tf.nn.bias_add(var, b)

def deconv2d(name, tensor, ksize, outshape, stddev=0.01, stride=2, padding='SAME'):
    with tf.variable_scope(name):
        w = tf.get_variable('w'
 
, [ksize, ksize, outshape[-1], tensor.get_shape()[-1]], dtype=tf.float32,
                            initializer=tf.random_normal_initializer(stddev=stddev))
        var = tf.nn.conv2d_transpose(tensor, w, outshape, strides=[1, stride, stride, 1], padding=padding)
        b = tf.get_variable('b', [outshape[-1]], 'float32', initializer=tf.constant_initializer(0.01))
        return tf.nn.bias_add(var, b)

def fully_connected(name,value, output_shape):
    with tf.variable_scope(name, reuse=None) as scope:
        shape = value.get_shape().as_list()
        w = tf.get_variable('w', [shape[1], output_shape], dtype=tf.float32,
                                    initializer=tf.random_normal_initializer(stddev=0.01))
        b = tf.get_variable('b', [output_shape], dtype=tf.float32, initializer=tf.constant_initializer(0.0))

        return tf.matmul(value, w) + b

def relu(name, tensor):
    return tf.nn.relu(tensor, name)

def lrelu(name,x, leak=0.2):
    return tf.maximum(x, leak * x, name=name)


DEPTH = 28
OUTPUT_SIZE = 28
batch_size = 64
def Discriminator(name,inputs,reuse):
    with tf.variable_scope(name, reuse=reuse):
        output = tf.reshape(inputs, [-1, 28, 28, 1])
        output1 = conv2d('d_conv_1', output, ksize=5, out_dim=DEPTH)
        output2 = lrelu('d_lrelu_1', output1)

        output3 = conv2d('d_conv_2', output2, ksize=5, out_dim=2*DEPTH)
        output4 = lrelu('d_lrelu_2', output3)

        output5 = conv2d('d_conv_3', output4, ksize=5, out_dim=4*DEPTH)
        output6 = lrelu('d_lrelu_3', output5)

        # output7 = conv2d('d_conv_4', output6, ksize=5, out_dim=8*DEPTH)
        # output8 = lrelu('d_lrelu_4', output7)

        chanel = output6.get_shape().as_list()
        output9 = tf.reshape(output6, [batch_size, chanel[1]*chanel[2]*chanel[3]])
        output0 = fully_connected('d_fc', output9, 1)
        return output0


def generator(name, reuse=False):
    with tf.variable_scope(name, reuse=reuse):
        noise = tf.random_normal([batch_size, 128])#.astype('float32')

        noise = tf.reshape(noise, [batch_size, 128], 'noise')
        output = fully_connected('g_fc_1', noise, 2*2*8*DEPTH)
        output = tf.reshape(output, [batch_size, 2, 2, 8*DEPTH], 'g_conv')

        output = deconv2d('g_deconv_1', output, ksize=5, outshape=[batch_size, 4, 4, 4*DEPTH])
        output = tf.nn.relu(output)
        output = tf.reshape(output, [batch_size, 4, 4, 4*DEPTH])

        output = deconv2d('g_deconv_2', output, ksize=5, outshape=[batch_size, 7, 7, 2* DEPTH])
        output = tf.nn.relu(output)

        output = deconv2d('g_deconv_3', output, ksize=5, outshape=[batch_size, 14, 14, DEPTH])
        output = tf.nn.relu(output)

        output = deconv2d('g_deconv_4', output, ksize=5, outshape=[batch_size, OUTPUT_SIZE, OUTPUT_SIZE, 1])
        # output = tf.nn.relu(output)
        output = tf.nn.sigmoid(output)
        return tf.reshape(output,[-1,784])


def save_images(images, size, path):
    # 圖片歸一化
    img = (images + 1.0) / 2.0
    h, w = img.shape[1], img.shape[2]
    merge_img = np.zeros((h * size[0], w * size[1], 3))
    for idx, image in enumerate(images):
        i = idx % size[1]
        j = idx // size[1]
        merge_img[j * h:j * h + h, i * w:i * w + w, :] = image
    return scipy.misc.imsave(path, merge_img)


LAMBDA = 10
EPOCH = 40
def train():
    # print  os.getcwd()
    with tf.variable_scope(tf.get_variable_scope()):
        # real_data = tf.placeholder(dtype=tf.float32, shape=[-1, OUTPUT_SIZE*OUTPUT_SIZE*3])
        path = os.getcwd()
        data_dir = path + "/train.tfrecords"#準備使用自己的資料集
        # print data_dir
        '''獲得資料'''
        z = tf.placeholder(dtype=tf.float32, shape=[batch_size, 100])#build placeholder
        real_data = tf.placeholder(tf.float32, shape=[batch_size,784])

        with tf.variable_scope(tf.get_variable_scope()):
            fake_data = generator('gen',reuse=False)
            disc_real = Discriminator('dis_r',real_data,reuse=False)
            disc_fake = Discriminator('dis_r',fake_data,reuse=True)

        t_vars = tf.trainable_variables()
        d_vars = [var for var in t_vars if 'd_' in var.name]
        g_vars = [var for var in t_vars if 'g_' in var.name]

        '''計算損失'''
        gen_cost = tf.reduce_mean(disc_fake)
        disc_cost = -tf.reduce_mean(disc_fake) + tf.reduce_mean(disc_real)

        alpha = tf.random_uniform(
            shape=[batch_size, 1],minval=0.,maxval=1.)
        differences = fake_data - real_data
        interpolates = real_data + (alpha * differences)
        gradients = tf.gradients(Discriminator('dis_r',interpolates,reuse=True), [interpolates])[0]
        slopes = tf.sqrt(tf.reduce_sum(tf.square(gradients), reduction_indices=[1]))
        gradient_penalty = tf.reduce_mean((slopes - 1.) ** 2)
        disc_cost += LAMBDA * gradient_penalty

        with tf.variable_scope(tf.get_variable_scope(), reuse=None):
            gen_train_op = tf.train.AdamOptimizer(
                learning_rate=1e-4,beta1=0.5,beta2=0.9).minimize(gen_cost,var_list=g_vars)
            disc_train_op = tf.train.AdamOptimizer(
                learning_rate=1e-4,beta1=0.5,beta2=0.9).minimize(disc_cost,var_list=d_vars)

        saver = tf.train.Saver()

        # os.environ['CUDA_VISIBLE_DEVICES'] = str(0)#gpu環境
        # config = tf.ConfigProto()
        # config.gpu_options.per_process_gpu_memory_fraction = 0.5#呼叫50%GPU資源
        # sess = tf.InteractiveSession(config=config)
        sess = tf.InteractiveSession()
        coord = tf.train.Coordinator()
        threads = tf.train.start_queue_runners(sess=sess, coord=coord)
        if not os.path.exists('img'):
            os.mkdir('img')

        init = tf.global_variables_initializer()
        # init = tf.initialize_all_variables()
        sess.run(init)
        mnist = input_data.read_data_sets("data", one_hot=True)
        # mnist = mnist_data.read_data_sets("data", one_hot=True, reshape=False, validation_size=0)
        for epoch in range (1, EPOCH):
            idxs = 1000
            for iters in range(1, idxs):
                img, _ = mnist.train.next_batch(batch_size)

                # img2 = tf.reshape(img, [batch_size, 784])
                for x in range (0,5):
                    _, d_loss = sess.run([disc_train_op, disc_cost], feed_dict={real_data: img})
                _, g_loss = sess.run([gen_train_op, gen_cost])
                # print "fake_data:%5f disc_real:%5f disc_fake:%5f "%(tf.reduce_mean(fake_data)
                #                         ,tf.reduce_mean(disc_real),tf.reduce_mean(disc_fake))
                print("[%4d:%4d/%4d] d_loss: %.8f, g_loss: %.8f"%(epoch, iters, idxs, d_loss, g_loss))

            with tf.variable_scope(tf.get_variable_scope()):
                samples = generator('gen', reuse=True)
                samples = tf.reshape(samples, shape=[batch_size, 28,28,1])
                samples=sess.run(samples)
                save_images(samples, [8,8], os.getcwd()+'/img/'+'sample_%d_epoch.png' % (epoch))

            if epoch>=39:
                checkpoint_path = os.path.join(os.getcwd(),
                                               'my_wgan-gp.ckpt')
                saver.save(sess, checkpoint_path, global_step=epoch)
                print '*********    model saved    *********'

        coord.request_stop()
        coord.join(threads)
        sess.close()
if __name__ == '__main__':
    train()

生成結果：

第一個epoch生成結果

第39個epoch生成結果

實驗總結：一開始使用DCGAN做實驗，但是怎麼調都不收斂，dcgan需要小心的平衡生成器和辨別器的訓練成都，中間換了好幾個學習率，效果都不太理想，就使用了wgan-gp，後者就好訓練多了，完全不用擔心訓練失衡的問題，用著還是很順手的。