不要怂，就是GAN (生成式对抗网络) （五）：无约束条件的 GAN 代码与网络的 Graph

GAN 这个领域发展太快，日新月异，各种 GAN 层出不穷，前几天看到一篇关于 Wasserstein GAN 的文章，讲的很好，在此把它分享出来一起学习：https://zhuanlan.zhihu.com/p/25071913。相比 Wasserstein GAN ，我们的 DCGAN 好像低了一个档次，但是我们伟大的教育家鲁迅先生说过：“合抱之木，生于毫末；九层之台，起于累土；千里之行，始于足下”，（依稀记得那大概是我 7 - 8 岁的时候，鲁迅先生依偎在我身旁，带着和蔼可亲切的口吻对我说的这句话，他当时还加了一句话，小伙子你要记住，如果一句名言，你不知道是谁说的，那就是鲁迅说的）。所以我们的基础还是要打好的， DCGAN 是我们的基础，有了 DCGAN 的代码经验，相信写起 Wasserstein GAN 就顺手很多，所以，我们接下来继续来研究我们的无约束条件 DCGAN。

在上一篇文章中，我们用 MNIST 手写字符训练 GAN，生成网络 G 生成了相对比较好的手写字符，这一次，我们换个数据集，用 CelebA 人脸数据集来训练我们的 GAN，相比于手写字符，人脸数据集的分布更加复杂多样，长头发短头发，黄种人黑种人，戴眼镜不戴眼镜，男人女人等等，看看我们的生成网络 G 能否成功的检验出人脸数据集的分布。

首先准备数据：从官网分享的百度云盘连接 https://pan.baidu.com/s/1eSNpdRG#list/path=%2FCelebA%2FImg 下载 img_align_celeba.zip，在 /home/your_name/TensorFlow/DCGAN/data 文件夹下解压，得到 img_align_celeba 文件夹，里面有 20600 张人脸图片，在 /home/your_name/TensorFlow/DCGAN/data 文件夹下新建 img_align_celeba_tfrecords 文件夹，用来存放 tfrecords 文件，然后，在 /home/your_name/TensorFlow/DCGAN/ 下新建 convert_data.py，编写如下的代码，把人脸图片转化成 tfrecords 形式：

import os
import time
from PIL import Image
 
import tensorflow as tf
 
# 将图片裁剪为 128 x 128
OUTPUT_SIZE = 128
# 图片通道数，3 表示彩色
DEPTH = 3
 
def _int64_feature(value):
    return tf.train.Feature(int64_list = tf.train.Int64List(value = [value]))
def _bytes_feature(value):
    return tf.train.Feature(bytes_list = tf.train.BytesList(value = [value]))
 
def convert_to(data_path, name):
 
    """
    Converts s dataset to tfrecords
    """
 
    rows = 64
    cols = 64
    depth = DEPTH
    # 循环 12 次，产生 12 个 .tfrecords 文件
    for ii in range(12):
        writer = tf.python_io.TFRecordWriter(name + str(ii) + '.tfrecords')
        # 每个 tfrecord 文件有 16384 个图片
        for img_name in os.listdir(data_path)[ii*16384 : (ii+1)*16384]:
            # 打开图片
            img_path = data_path + img_name
            img = Image.open(img_path)
            # 设置裁剪参数
            h, w = img.size[:2]
            j, k = (h - OUTPUT_SIZE) / 2, (w - OUTPUT_SIZE) / 2
            box = (j, k, j + OUTPUT_SIZE, k+ OUTPUT_SIZE)
            # 裁剪图片
            img = img.crop(box = box)
            # image resize
            img = img.resize((rows,cols))
            # 转化为字节
            img_raw = img.tobytes()
            # 写入到 Example
            example = tf.train.Example(features = tf.train.Features(feature = {
                                        'height': _int64_feature(rows),
                                        'width': _int64_feature(cols),
                                        'depth': _int64_feature(depth),
                                        'image_raw': _bytes_feature(img_raw)}))
            writer.write(example.SerializeToString())
        writer.close()
 
if __name__ == '__main__':
 
    current_dir = os.getcwd()
    data_path = current_dir + '/data/img_align_celeba/'
    name = current_dir + '/data/img_align_celeba_tfrecords/train'
    start_time = time.time() 
 
    print('Convert start')
    print('\n' * 2)
 
    convert_to(data_path, name)
 
    print('\n' * 2)
    print('Convert done, take %.2f seconds' % (time.time() - start_time))

运行之后，在 /home/your_name/TensorFlow/DCGAN/data/img_align_celeba_tfrecords/ 下会产生 12 个 .tfrecords 文件，这就是我们要的数据格式。

数据准备好之后，根据前面的经验，我们来写无约束条件的 DCGAN 代码，在 /home/your_name/TensorFlow/DCGAN/ 新建 none_cond_DCGAN.py 文件敲写代码，为了简便起见，代码中没有加注释并且把所有的代码总结到一个代码中，从代码中可以看到，我们自己写了一个 batch_norm 层，解决了 evaluation 函数中 is_train = False 的问题，并且可以断点续训练(只需要将开头的 LOAD_MODEL 设置为 True)；此外该程序在开头采用很多的宏定义，可以方便的改为 tf.app.flags 定义的命令行参数，进而在命令行终端进行训练，还可以进行类的拓展，例如：

class DCGAN(object):
    def __init__(self):
        self.BATCH_SIZE = 64
        ...
    def bias(self):
        ...
 
    ...

关于类的拓展，这里不做过多说明。

在 none_cond_DCGAN.py 文件中敲写如下代码：

import os
import numpy as np
import scipy.misc
import tensorflow as tf
 
BATCH_SIZE = 64
OUTPUT_SIZE = 64
GF = 64             # Dimension of G filters in first conv layer. default [64]
DF = 64             # Dimension of D filters in first conv layer. default [64]
Z_DIM = 100
IMAGE_CHANNEL = 3
LR = 0.0002         # Learning rate
EPOCH = 5
LOAD_MODEL = False  # Whether or not continue train from saved model。
TRAIN = True
CURRENT_DIR = os.getcwd()
 
def bias(name, shape, bias_start = 0.0, trainable = True):
 
    dtype = tf.float32
    var = tf.get_variable(name, shape, tf.float32, trainable = trainable,
                          initializer = tf.constant_initializer(
                                                  bias_start, dtype = dtype))
    return var
 
def weight(name, shape, stddev = 0.02, trainable = True):
 
    dtype = tf.float32
    var = tf.get_variable(name, shape, tf.float32, trainable = trainable,
                          initializer = tf.random_normal_initializer(
                                              stddev = stddev, dtype = dtype))
    return var
 
def fully_connected(value, output_shape, name = 'fully_connected', with_w = False):
 
    shape = value.get_shape().as_list()
 
    with tf.variable_scope(name):
        weights = weight('weights', [shape[1], output_shape], 0.02)
        biases = bias('biases', [output_shape], 0.0)
 
    if with_w:
        return tf.matmul(value, weights) + biases, weights, biases
    else:
        return tf.matmul(value, weights) + biases
 
def lrelu(x, leak=0.2, name = 'lrelu'):
 
    with tf.variable_scope(name):
        return tf.maximum(x, leak*x, name = name)
 
def relu(value, name = 'relu'):
    with tf.variable_scope(name):
        return tf.nn.relu(value)
 
def deconv2d(value, output_shape, k_h = 5, k_w = 5, strides =[1, 2, 2, 1],
             name = 'deconv2d', with_w = False):
 
    with tf.variable_scope(name):
        weights = weight('weights',
                         [k_h, k_w, output_shape[-1], value.get_shape()[-1]])
        deconv = tf.nn.conv2d_transpose(value, weights,
                                        output_shape, strides = strides)
        biases = bias('biases', [output_shape[-1]])
        deconv = tf.reshape(tf.nn.bias_add(deconv, biases), deconv.get_shape())
        if with_w:
            return deconv, weights, biases
        else:
            return deconv
 
def conv2d(value, output_dim, k_h = 5, k_w = 5,
            strides =[1, 2, 2, 1], name = 'conv2d'):
 
    with tf.variable_scope(name):
        weights = weight('weights',
                         [k_h, k_w, value.get_shape()[-1], output_dim])
        conv = tf.nn.conv2d(value, weights, strides = strides, padding = 'SAME')
        biases = bias('biases', [output_dim])
        conv = tf.reshape(tf.nn.bias_add(conv, biases), conv.get_shape())
 
        return conv
 
def conv_cond_concat(value, cond, name = 'concat'):
 
    """
    Concatenate conditioning vector on feature map axis.
    """
    value_shapes = value.get_shape().as_list()
    cond_shapes = cond.get_shape().as_list()
 
    with tf.variable_scope(name):
        return tf.concat(3,
                 [value, cond * tf.ones(value_shapes[0:3] + cond_shapes[3:])])
 
def batch_norm(value, is_train = True, name = 'batch_norm',
               epsilon = 1e-5, momentum = 0.9):
 
    with tf.variable_scope(name):
 
        ema = tf.train.ExponentialMovingAverage(decay = momentum)
        shape = value.get_shape().as_list()[-1]
        beta = bias('beta', [shape], bias_start = 0.0)
        gamma = bias('gamma', [shape], bias_start = 1.0)
 
        if is_train:
 
            batch_mean, batch_variance = tf.nn.moments(
                value, [0, 1, 2], name = 'moments')
 
            moving_mean = bias('moving_mean', [shape], 0.0, False)
            moving_variance = bias('moving_variance', [shape], 1.0, False)
 
            ema_apply_op = ema.apply([batch_mean, batch_variance])
 
            assign_mean = moving_mean.assign(ema.average(batch_mean))
            assign_variance = \
                moving_variance.assign(ema.average(batch_variance))
 
            with tf.control_dependencies([ema_apply_op]):
                mean, variance = \
                    tf.identity(batch_mean), tf.identity(batch_variance)
 
            with tf.control_dependencies([assign_mean, assign_variance]):
                return tf.nn.batch_normalization(
                    value, mean, variance, beta, gamma, 1e-5)
 
        else:
            mean = bias('moving_mean', [shape], 0.0, False)
            variance = bias('moving_variance', [shape], 1.0, False)
 
            return tf.nn.batch_normalization(
                value, mean, variance, beta, gamma, epsilon)
 
def generator(z, is_train = True, name = 'generator'):
 
    with tf.name_scope(name):
 
        s2, s4, s8, s16 = \
                    OUTPUT_SIZE/2, OUTPUT_SIZE/4, OUTPUT_SIZE/8, OUTPUT_SIZE/16
 
        h1 = tf.reshape(fully_connected(z, GF*8*s16*s16, 'g_fc1'),
                        [-1, s16, s16, GF*8], name = 'reshap')
        h1 = relu(batch_norm(h1, name = 'g_bn1', is_train = is_train))
 
        h2 = deconv2d(h1, [BATCH_SIZE, s8, s8, GF*4], name = 'g_deconv2d1')
        h2 = relu(batch_norm(h2, name = 'g_bn2', is_train = is_train))
 
        h3 = deconv2d(h2, [BATCH_SIZE, s4, s4, GF*2], name = 'g_deconv2d2')
        h3 = relu(batch_norm(h3, name = 'g_bn3', is_train = is_train))
 
        h4 = deconv2d(h3, [BATCH_SIZE, s2, s2, GF*1], name = 'g_deconv2d3')
        h4 = relu(batch_norm(h4, name = 'g_bn4', is_train = is_train))
 
        h5 = deconv2d(h4, [BATCH_SIZE, OUTPUT_SIZE, OUTPUT_SIZE, 3],
                      name = 'g_deconv2d4')    
 
        return tf.nn.tanh(h5)
 
def discriminator(image, reuse = False, name = 'discriminator'):
 
    with tf.name_scope(name):    
 
        if reuse:
            tf.get_variable_scope().reuse_variables()
 
        h0 = lrelu(conv2d(image, DF, name='d_h0_conv'), name = 'd_h0_lrelu')
        h1 = lrelu(batch_norm(conv2d(h0, DF*2, name='d_h1_conv'),
                              name = 'd_h1_bn'), name = 'd_h1_lrelu')
        h2 = lrelu(batch_norm(conv2d(h1, DF*4, name='d_h2_conv'),
                              name = 'd_h2_bn'), name = 'd_h2_lrelu')
        h3 = lrelu(batch_norm(conv2d(h2, DF*8, name='d_h3_conv'),
                              name = 'd_h3_bn'), name = 'd_h3_lrelu')
        h4 = fully_connected(tf.reshape(h3, [BATCH_SIZE, -1]), 1, 'd_h4_fc')
 
        return tf.nn.sigmoid(h4), h4
 
def sampler(z, is_train = False, name = 'sampler'):
 
    with tf.name_scope(name):
 
        tf.get_variable_scope().reuse_variables()
        return generator(z, is_train = is_train)
 
def read_and_decode(filename_queue):
 
    """
    read and decode tfrecords
    """
 
    reader = tf.TFRecordReader()
    _, serialized_example = reader.read(filename_queue)
 
    features = tf.parse_single_example(serialized_example,features = {
                        'image_raw':tf.FixedLenFeature([], tf.string)})
    image = tf.decode_raw(features['image_raw'], tf.uint8)
 
    image = tf.reshape(image, [OUTPUT_SIZE, OUTPUT_SIZE, 3])
    image = tf.cast(image, tf.float32)
    image = image / 255.0
 
    return image
 
def inputs(data_dir, batch_size, name = 'input'):
 
    """
    Reads input data num_epochs times.
    """
 
    with tf.name_scope(name):
        filenames = [
            os.path.join(data_dir,'train%d.tfrecords' % ii) for ii in range(12)]
        filename_queue = tf.train.string_input_producer(filenames)
 
        image = read_and_decode(filename_queue)
 
        images = tf.train.shuffle_batch([image], batch_size = batch_size,
                                        num_threads = 4,
                                        capacity = 20000 + 3 * batch_size,
                                        min_after_dequeue = 20000)
        return images
 
def save_images(images, size, path):
 
    """
    Save the samples images
    The best size number is
            int(max(sqrt(image.shape[1]),sqrt(image.shape[1]))) + 1
    """
    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)    
 
def train():
 
    global_step = tf.Variable(0, name = 'global_step', trainable = False)
 
    train_dir = CURRENT_DIR + '/logs_without_condition/'
    data_dir = CURRENT_DIR + '/data/img_align_celeba_tfrecords/'
 
    images = inputs(data_dir, BATCH_SIZE)
 
    z = tf.placeholder(tf.float32, [None, Z_DIM], name='z')
 
    G = generator(z)
    D, D_logits  = discriminator(images)
    samples = sampler(z)
    D_, D_logits_ = discriminator(G, reuse = True)
 
    d_loss_real = tf.reduce_mean(
            tf.nn.sigmoid_cross_entropy_with_logits(D_logits, tf.ones_like(D)))
    d_loss_fake = tf.reduce_mean(
            tf.nn.sigmoid_cross_entropy_with_logits(D_logits_, tf.zeros_like(D_)))
    d_loss = d_loss_real + d_loss_fake
    g_loss = tf.reduce_mean(
            tf.nn.sigmoid_cross_entropy_with_logits(D_logits_, tf.ones_like(D_)))
 
    z_sum = tf.histogram_summary('z', z)
    d_sum = tf.histogram_summary('d', D)
    d__sum = tf.histogram_summary('d_', D_)
    G_sum = tf.image_summary('G', G)
 
    d_loss_real_sum = tf.scalar_summary('d_loss_real', d_loss_real)
    d_loss_fake_sum = tf.scalar_summary('d_loss_fake', d_loss_fake)
    d_loss_sum = tf.scalar_summary('d_loss', d_loss)
    g_loss_sum = tf.scalar_summary('g_loss', g_loss)
 
    g_sum = tf.merge_summary([z_sum, d__sum, G_sum, d_loss_fake_sum, g_loss_sum])
    d_sum = tf.merge_summary([z_sum, d_sum, d_loss_real_sum, d_loss_sum])
 
    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]
 
    saver = tf.train.Saver()
 
    d_optim = tf.train.AdamOptimizer(LR, beta1 = 0.5) \
                .minimize(d_loss, var_list = d_vars, global_step = global_step)
    g_optim = tf.train.AdamOptimizer(LR, beta1 = 0.5) \
                .minimize(g_loss, var_list = g_vars, global_step = global_step)
 
    os.environ['CUDA_VISIBLE_DEVICES'] = str(0)
    config = tf.ConfigProto()
    config.gpu_options.per_process_gpu_memory_fraction = 0.2
    sess = tf.InteractiveSession(config=config)
 
    writer = tf.train.SummaryWriter(train_dir, sess.graph)    
 
    sample_z = np.random.uniform(-1, 1, size = (BATCH_SIZE, Z_DIM))
 
    coord = tf.train.Coordinator()
    threads = tf.train.start_queue_runners(sess = sess, coord = coord)
    init = tf.initialize_all_variables()
    sess.run(init)
 
    start = 0
    if LOAD_MODEL:
        print(" [*] Reading checkpoints...")
        ckpt = tf.train.get_checkpoint_state(train_dir)        
 
        if ckpt and ckpt.model_checkpoint_path:
            ckpt_name = os.path.basename(ckpt.model_checkpoint_path)
            saver.restore(sess, os.path.join(train_dir, ckpt_name))
            global_step = ckpt.model_checkpoint_path.split('/')[-1]\
                                                    .split('-')[-1]
            print('Loading success, global_step is %s' % global_step)
 
        start = int(global_step)
 
    for epoch in range(EPOCH):
 
        batch_idxs = 3072
 
        if epoch:
            start = 0
 
        for idx in range(start, batch_idxs):
 
            batch_z = np.random.uniform(-1, 1, size = (BATCH_SIZE, Z_DIM))
 
            _, summary_str = sess.run([d_optim, d_sum], feed_dict = {z: batch_z})
            writer.add_summary(summary_str, idx+1)
 
            # Update G network
            _, summary_str = sess.run([g_optim, g_sum], feed_dict = {z: batch_z})
            writer.add_summary(summary_str, idx+1)
 
            # Run g_optim twice to make sure that d_loss does not go to zero
            _, summary_str = sess.run([g_optim, g_sum], feed_dict = {z: batch_z})
            writer.add_summary(summary_str, idx+1)
 
            errD_fake = d_loss_fake.eval({z: batch_z})
            errD_real = d_loss_real.eval()
            errG = g_loss.eval({z: batch_z})
            if idx % 20 == 0:
                print("[%4d/%4d] d_loss: %.8f, g_loss: %.8f" \
                            % (idx, batch_idxs, errD_fake+errD_real, errG))
 
            if idx % 100 == 0:
                sample = sess.run(samples, feed_dict = {z: sample_z})
                samples_path = CURRENT_DIR + '/samples_without_condition/'
                save_images(sample, [8, 8],
                            samples_path + \
                            'sample_%d_epoch_%d.png' % (epoch, idx))
 
                print '\n'*2
                print('===========    %d_epoch_%d.png save down    ==========='
                                                                %(epoch, idx))
                print '\n'*2
 
            if (idx % 512 == 0) or (idx + 1 == batch_idxs):
                checkpoint_path = os.path.join(train_dir,
                                               'my_dcgan_tfrecords.ckpt')
                saver.save(sess, checkpoint_path, global_step = idx+1)
                print '*********    model saved    *********'
 
        print '******* start with %d *******' % start
 
    coord.request_stop()
    coord.join(threads)
    sess.close()
 
def evaluate():
    eval_dir = CURRENT_DIR + '/eval/'
 
    checkpoint_dir = CURRENT_DIR + '/logs_without_condition/'
 
    z = tf.placeholder(tf.float32, [None, Z_DIM], name='z')
 
    G = generator(z, is_train = False)
 
    sample_z1 = np.random.uniform(-1, 1, size=(BATCH_SIZE, Z_DIM))
    sample_z2 = np.random.uniform(-1, 1, size=(BATCH_SIZE, Z_DIM))
    sample_z3 = (sample_z1 + sample_z2) / 2
    sample_z4 = (sample_z1 + sample_z3) / 2
    sample_z5 = (sample_z2 + sample_z3) / 2    
 
    print("Reading checkpoints...")
    ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
 
    saver = tf.train.Saver(tf.all_variables())
 
    os.environ['CUDA_VISIBLE_DEVICES'] = str(0)
    config = tf.ConfigProto()
    config.gpu_options.per_process_gpu_memory_fraction = 0.2
    sess = tf.InteractiveSession(config=config)
 
    if ckpt and ckpt.model_checkpoint_path:
        ckpt_name = os.path.basename(ckpt.model_checkpoint_path)
        global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
        saver.restore(sess, os.path.join(checkpoint_dir, ckpt_name))
        print('Loading success, global_step is %s' % global_step)
 
    eval_sess1 = sess.run(G, feed_dict = {z: sample_z1})
    eval_sess2 = sess.run(G, feed_dict = {z: sample_z4})
    eval_sess3 = sess.run(G, feed_dict = {z: sample_z3})
    eval_sess4 = sess.run(G, feed_dict = {z: sample_z5})
    eval_sess5 = sess.run(G, feed_dict = {z: sample_z2})
 
    print(eval_sess3.shape)
 
    save_images(eval_sess1, [8, 8], eval_dir + 'eval_%d.png' % 1)
    save_images(eval_sess2, [8, 8], eval_dir + 'eval_%d.png' % 2)
    save_images(eval_sess3, [8, 8], eval_dir + 'eval_%d.png' % 3)
    save_images(eval_sess4, [8, 8], eval_dir + 'eval_%d.png' % 4)
    save_images(eval_sess5, [8, 8], eval_dir + 'eval_%d.png' % 5)
 
    sess.close()
 
if __name__ == '__main__':
 
    if TRAIN:
        train()
    else:
        evaluate()

完成后，运行代码，网络开始训练，大致需要 1~2 个小时，训练就可以完成，在训练的过程中，可以看出 sampler 采样的生成结果越来越好，最后得到了一个如下图所示的结果，由于人脸的数据分布比手写数据分布复杂多样，所以生成器不能完全抓住人脸的特征，下图所示的第 6 行第 7 列就是一个很糟糕的生成图像。

训练完成后，我们用 tensorboard 打开网络的 graph，看看经过我们的精心设计，网络结构变成了什么样子：

可以看出来，这次的结构图，比之前的顺眼多了，简直是处女座的福音啊有木有。

至此，我们完成了 DCGAN 的代码，下一篇文章，我们来说说 Caffe 那点事。

参考文献：

1. https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/how_tos/reading_data/convert_to_records.py

2. https://github.com/carpedm20/DCGAN-tensorflow