GAN训练效果出现问题

网上现有的GAN实现手写数字生成的代码，代码未做任何改动，在自己电脑上面运行的效果却很差。这是什么原因呢？开始训练还算正常，后面就出现了下面的问题。正常是可以计算下去的。

img


图片逐渐失真

img


代码就是下面网上常见的代码。


```python
from __future__ import print_function, division
from keras.datasets import mnist
from keras.layers import Input, Dense, Reshape, Flatten, Dropout
from keras.layers import BatchNormalization, Activation, ZeroPadding2D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.convolutional import UpSampling2D, Conv2D
from keras.models import Sequential, Model
from tensorflow.keras.optimizers import Adam
import matplotlib.pyplot as plt
import numpy as np

class DCGAN():
    def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002)

        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        self.generator = self.build_generator()

        z = Input(shape=(self.latent_dim,))
        img = self.generator(z)

        self.discriminator.trainable = False

        valid = self.discriminator(img)

        self.combined = Model(z, valid)
        self.combined.compile(loss='binary_crossentropy', optimizer=optimizer)

    def build_generator(self):

        model = Sequential()

        model.add(Dense(128 * 7 * 7, activation="relu", input_dim=self.latent_dim))
        model.add(Reshape((7, 7, 128)))
        model.add(UpSampling2D())
        model.add(Conv2D(128, kernel_size=3, padding="same"))
        model.add(BatchNormalization(momentum=0.8))
        model.add(Activation("relu"))
        model.add(UpSampling2D())
        model.add(Conv2D(64, kernel_size=3, padding="same"))
        model.add(BatchNormalization(momentum=0.8))
        model.add(Activation("relu"))
        model.add(Conv2D(self.channels, kernel_size=3, padding="same"))
        model.add(Activation("tanh"))

        model.summary()

        noise = Input(shape=(self.latent_dim,))
        img = model(noise)

        return Model(noise, img)

    def build_discriminator(self):

        model = Sequential()

        model.add(Conv2D(32, kernel_size=3, strides=2, input_shape=self.img_shape, padding="same"))
        model.add(LeakyReLU(alpha=0.2))
        model.add(Dropout(0.25))
        model.add(Conv2D(64, kernel_size=3, strides=2, padding="same"))
        model.add(ZeroPadding2D(padding=((0,1),(0,1))))
        model.add(BatchNormalization(momentum=0.8))
        model.add(LeakyReLU(alpha=0.2))
        model.add(Dropout(0.25))
        model.add(Conv2D(128, kernel_size=3, strides=2, padding="same"))
        model.add(BatchNormalization(momentum=0.8))
        model.add(LeakyReLU(alpha=0.2))
        model.add(Dropout(0.25))
        model.add(Conv2D(256, kernel_size=3, strides=1, padding="same"))
        model.add(BatchNormalization(momentum=0.8))
        model.add(LeakyReLU(alpha=0.2))
        model.add(Dropout(0.25))
        model.add(Flatten())
        model.add(Dense(1, activation='sigmoid'))

        model.summary()

        img = Input(shape=self.img_shape)
        validity = model(img)

        return Model(img, validity)

    def train(self, iter, batch_size=128, save_interval=50):

        (X_train, _), (_, _) = mnist.load_data()

        X_train = X_train / 127.5 - 1.
        X_train = np.expand_dims(X_train, axis=3)

        valid = np.ones((batch_size, 1))
        fake = np.zeros((batch_size, 1))

        for i in range(iter):

            idx = np.random.randint(0, X_train.shape[0], batch_size)
            imgs = X_train[idx]

            noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
            gen_imgs = self.generator.predict(noise)

            d_loss_real = self.discriminator.train_on_batch(imgs, valid)
            d_loss_fake = self.discriminator.train_on_batch(gen_imgs, fake)
            d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)

            g_loss = self.combined.train_on_batch(noise, valid)

            print ("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (i, d_loss[0], 100*d_loss[1], g_loss))

            if i % save_interval == 0:
                self.save_imgs(i)

    def save_imgs(self, iter):
        r, c = 5, 5
        noise = np.random.normal(0, 1, (r * c, self.latent_dim))
        gen_imgs = self.generator.predict(noise)

        gen_imgs = 0.5 * gen_imgs + 0.5

        fig, axs = plt.subplots(r, c)
        cnt = 0
        for i in range(r):
            for j in range(c):
                axs[i,j].imshow(gen_imgs[cnt, :,:,0], cmap='gray')
                axs[i,j].axis('off')
                cnt += 1
        fig.savefig("n/mnist_%d.png" % iter)
        plt.close()


if __name__ == '__main__':
    dcgan = DCGAN()
    dcgan.train(iter=10000, batch_size=32, save_interval=50)

从你的图片来看，过拟合了应该，所以越来越差了
你可以将这几个返回值保存依赖，然后绘制一下图片看下

img

  • 建议你看下这篇博客👉 ：GAN生成对抗网络
  • 除此之外, 这篇博客: 对抗网络（GAN）手写数字生成中的 1.跑通代码 部分也许能够解决你的问题, 你可以仔细阅读以下内容或跳转源博客中阅读:

  • 我这个人对于任何代码，我都会先去跑通之和才会去观看内容，哈哈哈，所以第一步我们先不管37=21，直接把博主的代码复制黏贴一份运行结果。（PS：做了一些修改，因为原文是jupyter，而我在pycharm）

    import tensorflow as tf

gpus = tf.config.list_physical_devices("GPU")

if gpus:
    tf.config.experimental.set_memory_growth(gpus[0], True)  # 设置GPU显存用量按需使用
    tf.config.set_visible_devices([gpus[0]], "GPU")

# 打印显卡信息，确认GPU可用
print(gpus)



from tensorflow.keras import layers, datasets, Sequential, Model, optimizers
from tensorflow.keras.layers import LeakyReLU, UpSampling2D, Conv2D

import matplotlib.pyplot as plt
import numpy             as np
import sys,os,pathlib


img_shape  = (28, 28, 1)
latent_dim = 200


def build_generator():
    # ======================================= #
    #     生成器，输入一串随机数字生成图片
    # ======================================= #
    model = Sequential([
        layers.Dense(256, input_dim=latent_dim),
        layers.LeakyReLU(alpha=0.2),  # 高级一点的激活函数
        layers.BatchNormalization(momentum=0.8),  # BN 归一化

        layers.Dense(512),
        layers.LeakyReLU(alpha=0.2),
        layers.BatchNormalization(momentum=0.8),

        layers.Dense(1024),
        layers.LeakyReLU(alpha=0.2),
        layers.BatchNormalization(momentum=0.8),

        layers.Dense(np.prod(img_shape), activation='tanh'),
        layers.Reshape(img_shape)
    ])

    noise = layers.Input(shape=(latent_dim,))
    img = model(noise)

    return Model(noise, img)

def build_discriminator():
    # ===================================== #
    #   鉴别器，对输入的图片进行判别真假
    # ===================================== #
    model = Sequential([
        layers.Flatten(input_shape=img_shape),
        layers.Dense(512),
        layers.LeakyReLU(alpha=0.2),
        layers.Dense(256),
        layers.LeakyReLU(alpha=0.2),
        layers.Dense(1, activation='sigmoid')
    ])

    img = layers.Input(shape=img_shape)
    validity = model(img)

    return Model(img, validity)

# 创建判别器
discriminator = build_discriminator()
# 定义优化器
optimizer = tf.keras.optimizers.Adam(1e-4)
discriminator.compile(loss='binary_crossentropy',
                      optimizer=optimizer,
                      metrics=['accuracy'])

# 创建生成器
generator = build_generator()
gan_input = layers.Input(shape=(latent_dim,))
img = generator(gan_input)

# 在训练generate的时候不训练discriminator
discriminator.trainable = False

# 对生成的假图片进行预测
validity = discriminator(img)
combined = Model(gan_input, validity)
combined.compile(loss='binary_crossentropy', optimizer=optimizer)

def sample_images(epoch):
    """
    保存样例图片
    """
    row, col = 4, 4
    noise = np.random.normal(0, 1, (row*col, latent_dim))
    gen_imgs = generator.predict(noise)

    fig, axs = plt.subplots(row, col)
    cnt = 0
    for i in range(row):
        for j in range(col):
            axs[i,j].imshow(gen_imgs[cnt, :,:,0], cmap='gray')
            axs[i,j].axis('off')
            cnt += 1
    fig.savefig("images/%05d.png" % epoch)
   # fig.savefig(" E:/2021_Project_YanYiXia/AI/21/对抗网络（GAN）手写数字生成/images/%05d.png" % epoch)

    plt.close()


def train(epochs, batch_size=128, sample_interval=50):
    # 加载数据
    (train_images, _), (_, _) = tf.keras.datasets.mnist.load_data()

    # 将图片标准化到 [-1, 1] 区间内
    train_images = (train_images - 127.5) / 127.5
    # 数据
    train_images = np.expand_dims(train_images, axis=3)

    # 创建标签
    true = np.ones((batch_size, 1))
    fake = np.zeros((batch_size, 1))

    # 进行循环训练
    for epoch in range(epochs):

        # 随机选择 batch_size 张图片
        idx = np.random.randint(0, train_images.shape[0], batch_size)
        imgs = train_images[idx]

        # 生成噪音
        noise = np.random.normal(0, 1, (batch_size, latent_dim))
        # 生成器通过噪音生成图片，gen_imgs的shape为：(128, 28, 28, 1)
        gen_imgs = generator.predict(noise)

        # 训练鉴别器
        d_loss_true = discriminator.train_on_batch(imgs, true)
        d_loss_fake = discriminator.train_on_batch(gen_imgs, fake)
        # 返回loss值
        d_loss = 0.5 * np.add(d_loss_true, d_loss_fake)

        # 训练生成器
        noise = np.random.normal(0, 1, (batch_size, latent_dim))
        g_loss = combined.train_on_batch(noise, true)

        print("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100 * d_loss[1], g_loss))

        # 保存样例图片
        if epoch % sample_interval == 0:
            sample_images(epoch)

#train(epochs=30000, batch_size=256, sample_interval=200)


import imageio


def compose_gif():
    # 图片地址
    data_dir = "E:/2021_Project_YanYiXia/AI/21/对抗网络（GAN）手写数字生成/images"
    data_dir = pathlib.Path(data_dir)
    paths = list(data_dir.glob('*'))

    gif_images = []
    for path in paths:
        print(path)
        gif_images.append(imageio.imread(path))
    imageio.mimsave("test.gif", gif_images, fps=2)


compose_gif()

    点击pycharm运行即可得到结果，此图为对抗网络生成的手写数字

  • 您还可以看一下 CSDN讲师老师的详解GAN在黑白照片上色中的应用课程中的 详解GAN在黑白照片上色中的应用-下小节, 巩固相关知识点