使用DCGAN（deep convolutional GAN）：深度卷积GAN

网络结构如下：

代码分成四个文件：

• 读入文件                                          read_data.py
• 配置线性层，卷积层，反卷积层     ops.py
• 构建生成器和判别器模型                model.py
• 训练模型                                          train.py

使用的layer种类有：

• conv（卷积层）
• deconv（反卷积层）
• linear（线性层）
• batch_norm（批量归一化层）
• lrelu/relu/sigmoid（非线性函数层）

一、读入文件（read_data.py）

import osimport numpy as npimport tensorflow as tfdef read_data():    data_dir = "data\mnist"    #read training data    fd = open(os.path.join(data_dir,"train-images.idx3-ubyte"))    loaded = np.fromfile(file = fd, dtype = np.uint8)    trainX = loaded[16:].reshape((60000, 28, 28, 1)).astype(np.float)    fd = open(os.path.join(data_dir,"train-labels.idx1-ubyte"))    loaded = np.fromfile(file = fd, dtype = np.uint8)    trainY = loaded[8:].reshape((60000)).astype(np.float)    #read test data    fd = open(os.path.join(data_dir,"t10k-images.idx3-ubyte"))    loaded = np.fromfile(file = fd, dtype = np.uint8)    testX = loaded[16:].reshape((10000, 28, 28, 1)).astype(np.float)    fd = open(os.path.join(data_dir,"t10k-labels.idx1-ubyte"))    loaded = np.fromfile(file = fd, dtype = np.uint8)    testY = loaded[8:].reshape((10000)).astype(np.float)    # 将两个集合合并成70000大小的数据集    X = np.concatenate((trainX, testX), axis = 0)    y = np.concatenate((trainY, testY), axis = 0)    print(X[:2])    #set the random seed    seed = 233    np.random.seed(seed)    np.random.shuffle(X)    np.random.seed(seed)    np.random.shuffle(y)    return X/255, y

新建一个data文件夹，存放mnist数据集。读入训练集和测试集，将两个集合合并成一个70000的数据集，设置相同的随机种子，保证两个集合随机成相同的顺序。最后把X归一化到 [0, 1] 之间

二、配置线性层，卷积层，反卷积层 （ops.py）

import tensorflow as tffrom tensorflow.contrib.layers.python.layers import batch_norm as batch_normdef linear_layer(value, output_dim, name = 'linear_connected'):    with tf.variable_scope(name):        try:            # 线性层的权重                # 名称：weights                # shape = [int(value.get_shape()[1]), output_dim], 我们需要传入最后输出多少维，也即是output_dim的值                # 初始化：使用标准正态截断函数 标准差是0.02            weights = tf.get_variable('weights',                [int(value.get_shape()[1]), output_dim],                initializer = tf.truncated_normal_initializer(stddev = 0.02))            # 线性层的偏置                # 名称：biases                # shape = [output_dim], 偏置与输出的维度相同                # 初始化：使用常量初始化  初始化为0            biases = tf.get_variable('biases',                [output_dim], initializer = tf.constant_initializer(0.0))        except ValueError:            tf.get_variable_scope().reuse_variables()            weights = tf.get_variable('weights',                [int(value.get_shape()[1]), output_dim],                initializer = tf.truncated_normal_initializer(stddev = 0.02))            biases = tf.get_variable('biases',                [output_dim], initializer = tf.constant_initializer(0.0))        return tf.matmul(value, weights) + biasesdef conv2d(value, output_dim, k_h = 5, k_w = 5, strides = [1,1,1,1], name = "conv2d"):    with tf.variable_scope(name):        try:            # 反卷积层的权重                # 名称：weights                # [5, 5, 输入的维度， 输出的维度]                # 初始化：使用标准正态截断函数 标准差是0.02            weights = tf.get_variable('weights',                [k_h, k_w, int(value.get_shape()[-1]), output_dim],                initializer = tf.truncated_normal_initializer(stddev = 0.02))            # 线性层的偏置                # 名称：biases                # shape = [output_shape[-1]], 偏置与输出的维度相同                # 初始化：使用常量初始化  初始化为0            biases = tf.get_variable('biases',                [output_dim], initializer = tf.constant_initializer(0.0))        except ValueError:            tf.get_variable_scope().reuse_variables()            weights = tf.get_variable('weights',                [k_h, k_w, int(value.get_shape()[-1]), output_dim],                initializer = tf.truncated_normal_initializer(stddev = 0.02))            biases = tf.get_variable('biases',                [output_dim], initializer = tf.constant_initializer(0.0))        conv = tf.nn.conv2d(value, weights, strides = strides, padding = "SAME")        conv = tf.reshape(tf.nn.bias_add(conv, biases), conv.get_shape())        return convdef deconv2d(value, output_shape, k_h = 5, k_w = 5, strides = [1,1,1,1], name = "deconv2d"):    with tf.variable_scope(name):        try:            # 反卷积层的权重            # filter : [height, width, output_channels, in_channels]                # 名称：weights                # [5, 5, 输出的维度， 输入的维度]                # 初始化：使用标准正态截断函数 标准差是0.02            weights = tf.get_variable('weights',                [k_h, k_w, output_shape[-1], int(value.get_shape()[-1])],                initializer = tf.truncated_normal_initializer(stddev = 0.02))            # 线性层的偏置                # 名称：biases                # shape = [output_shape[-1]], 偏置与输出的维度相同                # 初始化：使用常量初始化  初始化为0            biases = tf.get_variable('biases',                [output_shape[-1]], initializer = tf.constant_initializer(0.0))        except ValueError:            tf.get_variable_scope().reuse_variables()            weights = tf.get_variable('weights',                [k_h, k_w, output_shape[-1], int(value.get_shape()[-1])],                initializer = tf.truncated_normal_initializer(stddev = 0.02))            biases = tf.get_variable('biases',                [output_shape[-1]], initializer = tf.constant_initializer(0.0))        deconv = tf.nn.conv2d_transpose(value, weights, output_shape, strides = strides)        deconv = tf.reshape(tf.nn.bias_add(deconv, biases), deconv.get_shape())        return deconvdef conv_cond_concat(value, cond, name = 'concat'):    value_shapes = value.get_shape().as_list()    cond_shapes = cond.get_shape().as_list()    with tf.variable_scope(name):        return tf.concat([value, cond * tf.ones(value_shapes[0:3] + cond_shapes[3:])], 3, name = name)# 批量归一化层def batch_norm_layer(value, is_train = True, name = 'batch_norm'):    with tf.variable_scope(name) as scope:        if is_train:            return batch_norm(value, decay = 0.9, epsilon = 1e-5, scale = True,                                is_training = is_train, updates_collections = None, scope = scope)        else :            return batch_norm(value, decay = 0.9, epsilon = 1e-5, scale = True,                            is_training = is_train, reuse = True,                            updates_collections = None, scope = scope)def lrelu(x, leak = 0.2, name = 'lrelu'):    with tf.variable_scope(name):        return tf.maximum(x, x*leak, name = name)

主要就是配置各个层的权重和参数，conv_cond_concat是为了用于卷积层计算的四维数据[batch_size, w, h, c]和约束条件y连接起来的操作，需要把两个数据的前三维转化到一样大小才能使用tf.concat(拼接函数)

lrelu是relu的改良版。代码中a = 0.2

三、构建生成器和判别器模型（model.py）

import tensorflow as tffrom ops import *BATCH_SIZE = 64# 生成器模型def generator(z, y,train = True):    with tf.variable_scope('generator', reuse=tf.AUTO_REUSE):        # 生成器的输入，z是 1 * 100 维的噪声        yb = tf.reshape(y, [BATCH_SIZE, 1, 1, 10], name = 'g_yb')        # 将噪声 z 和 y 拼接起来        z_y = tf.concat([z,y], 1, name = 'g_z_concat_y')        # 经过线性层，z_y 变成1024维        linear1 = linear_layer(z_y, 1024, name = 'g_linear_layer1')        # 先批量归一化，然后经过relu层，得 bn1        bn1 = tf.nn.relu(batch_norm_layer(linear1, is_train = True, name = 'g_bn1'))        # 将 bn1 与 y 拼接在一起        bn1_y = tf.concat([bn1, y], 1 ,name = 'g_bn1_concat_y')        # 经过线性层，变成128 * 49        linear2 = linear_layer(bn1_y, 128*49, name = 'g_linear_layer2')        # 先批量归一化，然后经过relu层，得 bn2        bn2 = tf.nn.relu(batch_norm_layer(linear2, is_train = True, name = 'g_bn2'))        # reshape成 7 * 7 * 128        bn2_re = tf.reshape(bn2, [BATCH_SIZE, 7, 7, 128], name = 'g_bn2_reshape')        # 将 bn2_re 与 yb 连接起来        bn2_yb = conv_cond_concat(bn2_re, yb, name = 'g_bn2_concat_yb')        # 经过反卷积 步长设为2，height和width 都翻倍        deconv1 = deconv2d(bn2_yb, [BATCH_SIZE, 14, 14, 128], strides = [1, 2, 2, 1], name = 'g_deconv1')        # 先批量归一化，然后经过relu层，得 bn3        bn3 = tf.nn.relu(batch_norm_layer(deconv1, is_train = True, name = 'g_bn3'))        # 将 bn2_re 与 yb 连接起来        bn3_yb = conv_cond_concat(bn3, yb, name = 'g_bn3_concat_yb')        # 经过反卷积 步长设为2，height和width 都翻倍        deconv2 = deconv2d(bn3_yb, [BATCH_SIZE, 28, 28, 1], strides = [1, 2, 2, 1], name = 'g_deconv2')        # 经过sigmoid层        return tf.nn.sigmoid(deconv2)# 判别器模型def discriminator(image, y, reuse=tf.AUTO_REUSE, train = True):    with tf.variable_scope('discriminator', reuse=tf.AUTO_REUSE):        # 判别器有两个输入，一个是真实图片，另一个是生成的图片，都是 28 * 28        yb = tf.reshape(y, [BATCH_SIZE, 1, 1, 10], name = 'd_yb')        # 将 image 与 yb 连接起来        image_yb = conv_cond_concat(image, yb, name = 'd_image_concat_yb')        # 卷积：步长为2   14 * 14 * 11        conv1 = conv2d(image_yb, 11, strides = [1, 2, 2, 1], name = 'd_conv1')        # 经过 lrelu 层        lr1 = lrelu(conv1, name = 'd_lrelu1')        # 将 lr1 与 yb 连接起来        lr1_yb = conv_cond_concat(lr1, yb, name = 'd_lr1_concat_yb')        # 卷积：步长为2   7 * 7 * 74        conv2 = conv2d(lr1_yb, 74, strides = [1, 2, 2, 1], name = 'd_conv2')        # 批量归一化        bn1 = batch_norm_layer(conv2, is_train = True, name = 'd_bn1')        # 经过 lrelu 层        lr2 = lrelu(bn1, name = 'd_lrelu2')        lr2_re = tf.reshape(lr2, [BATCH_SIZE, -1], name = 'd_lr2_reshape')        # 将 lr2_re 与 y 连接起来        lr2_y = tf.concat([lr2_re, y], 1, name = 'd_lr2_concat_y')        # 经过线性层 变成 1 * 1024        linear1 = linear_layer(lr2_y, 1024, name = 'd_linear_layer1')        # 批量归一化        bn2 = batch_norm_layer(linear1, is_train = True, name = 'd_bn2')        # 经过 lrelu 层        lr3 = lrelu(bn2, name = 'd_lrelu3')        lr3_y = tf.concat([lr3, y], 1, name = 'd_lr3_concat_y')        linear2 = linear_layer(lr3_y, 1, name = 'd_linear_layer2')        return linear2def sampler(z, y, train = True):    tf.get_variable_scope().reuse_variables()    return generator(z, y, train = train)

按照模型图实现生成器G，返回时用到了sigmoid，将输出值规范到（0， 1）与前面输入图像的范围一致。

判别器D：

1、 设置reuse变量（变量重复使用）。判别器有两个输入。对于同一个D，先喂给它real data（真实的图像），然后喂给它fake data（生成器生成的假图像）。在一次train_step里涉及到了两次D变量的重复使用，所以需要设置变量共享。

2、返回值没有使用sigmoid，因为训练中使用了sigmoid_cross_entropy_with_logits来计算loss

四、训练模型

import scipy.miscimport numpy as npimport tensorflow as tfimport osfrom read_data import *from ops import *from model import *BATCH_SIZE = 64# 保存图片，将图片拼接在一起def save_images(images, size, path):    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():    #read data    X, Y = read_data()    #global_step to record the step of training    global_step = tf.Variable(0, name = 'global_step', trainable = True)    # 创建placeholder    y = tf.placeholder(tf.int32, [BATCH_SIZE], name = 'y')    _y = tf.one_hot(y, depth = 10, on_value=None, off_value=None, axis=None, dtype=None, name='one_hot')    z = tf.placeholder(tf.float32, [BATCH_SIZE, 100], name = 'z')    images = tf.placeholder(tf.float32, [BATCH_SIZE, 28, 28, 1], name = 'images')    # 用噪声生成图片    G = generator(z, _y)    # 用真实图片训练判别器    D = discriminator(images, _y)    # 用生成的加图片G训练判别器    _D = discriminator(G, _y)    # 计算损失值，使用 sigmoid_cross_entropy_with_logits    # 真实图片在判别器里产生的loss，我们希望真实图片在判别器里的label趋向于1    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = D, labels = tf.ones_like(D)))    # 假图片在判别器里产生的loss，我们希望假图片在判别器里的label趋向于0    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = _D, labels = tf.zeros_like(_D)))    # 假图片在判别器上的结果，我们希望它以假乱真，希望它尽可能接近于1    g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = _D, labels = tf.ones_like(_D)))        # 判别器的总 loss    d_loss = d_loss_real + d_loss_fake    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]    with tf.variable_scope(tf.get_variable_scope(), reuse = False):        # 优化判别器的 loss        d_optim = tf.train.AdamOptimizer(0.0002, beta1 = 0.5).minimize(d_loss, var_list = d_vars, global_step = global_step)        # 优化生成器的 loss        g_optim = tf.train.AdamOptimizer(0.0002, beta2 = 0.5).minimize(g_loss, var_list = g_vars, global_step = global_step)    #tensorborad    train_dir = 'logs'    z_sum = tf.summary.histogram("z",z)    d_sum = tf.summary.histogram("d",D)    d__sum = tf.summary.histogram("d_",_D)    g_sum = tf.summary.histogram("g", G)    d_loss_real_sum = tf.summary.scalar("d_loss_real", d_loss_real)    d_loss_fake_sum = tf.summary.scalar("d_loss_fake", d_loss_fake)    g_loss_sum = tf.summary.scalar("g_loss", g_loss)    d_loss_sum = tf.summary.scalar("d_loss", d_loss)    g_sum = tf.summary.merge([z_sum, d__sum, g_sum, d_loss_fake_sum, g_loss_sum])    d_sum = tf.summary.merge([z_sum, d_sum, d_loss_real_sum, d_loss_sum])    #initial    init = tf.global_variables_initializer()    sess = tf.InteractiveSession()    writer = tf.summary.FileWriter(train_dir, sess.graph)    #save    saver = tf.train.Saver()    # 保存模型    check_path = "./save/model.ckpt"    #sample    sample_z = np.random.uniform(-1, 1, size = (BATCH_SIZE, 100))    sample_labels = Y[0:BATCH_SIZE]    #make sample    sample = sampler(z, _y)    #run    sess.run(init)    #saver.restore(sess.check_path)    #train    for epoch in range(10):        batch_idx = int(70000/64)        for idx in range(batch_idx):            batch_images = X[idx*64:(idx+1)*64]            batch_labels = Y[idx*64:(idx+1)*64]            batch_z = np.random.uniform(-1, 1, size = (BATCH_SIZE, 100))            _, summary_str = sess.run([d_optim, d_sum],                                    feed_dict = {images: batch_images,                                                 z: batch_z,                                                 y: batch_labels})            writer.add_summary(summary_str, idx+1)            _, summary_str = sess.run([g_optim, g_sum],                                    feed_dict = {images: batch_images,                                                 z: batch_z,                                                 y: batch_labels})            writer.add_summary(summary_str, idx+1)            d_loss1 = d_loss_fake.eval({z: batch_z, y: batch_labels})            d_loss2 = d_loss_real.eval({images: batch_images, y:batch_labels})            D_loss = d_loss1 + d_loss2            G_loss = g_loss.eval({z: batch_z, y: batch_labels})            #every 20 batch output loss            if idx % 20 == 0:                print("Epoch: %d [%4d/%4d] d_loss: %.8f, g_loss: %.8f" % (epoch, idx, batch_idx, D_loss, G_loss))            #every 100 batch save a picture            if idx % 100 == 0:                sap = sess.run(sample, feed_dict = {z: sample_z, y: sample_labels})                samples_path = './sample/'                save_images(sap, [8,8], samples_path+'test_%d_epoch_%d.png' % (epoch, idx))            #every 500 batch save model            if idx % 500 == 0:                saver.save(sess, check_path, global_step = idx + 1)    sess.close()if __name__ == '__main__':    train()

模型的训练顺序是先generator生成fake data，然后real data喂给D训练，再把fake data喂给D训练

分别计算生成器和判别器的loss，其中判别器的loss = real loss + fake loss

一个batch中，训练一次D训练一次G。按理说应该训练k次D，训练一次G（为了方便就没有这么做了）

生成的效果图如下：