BiGAN 简介与代码实战

1.介绍
  在AAEGan（AAEGan 简介与代码实战）中，图片经过编码端可以被提取出隐变量z，隐变量z通过解码端可以被生成图片，其实这里面就有一种对称美，我们今天的猪脚BiGan也是有对称美。更加详细的内容可以参见论文： ADVERSARIAL FEATURE LEARNING

2.模型结构
  整个结构包括三部分：Encode网络，G网络，D网络。

   Encode网络，提取原始图片的隐变量

   G网络，将噪声生成图片

   D网络，判断这个配对（原始图片和隐变量      生成图片和噪声）是来自encoder还是decoder

 

3.模型特点
      bigan的全称是Bidirectional GAN，中文翻译是“双向Gan”，对于配对“原始图片和隐变量 ”原始图片是已知的，隐变量 是学习噪声，对于配对“生成图片和噪声”，噪声是已知的，生成图片是学习原始图片，这就是一个双向学习过程。

 4.代码实现keras

class BIGAN():
    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, 0.5)
 
        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss=['binary_crossentropy'],
            optimizer=optimizer,
            metrics=['accuracy'])
 
        # Build the generator
        self.generator = self.build_generator()
 
        # Build the encoder
        self.encoder = self.build_encoder()
 
        # The part of the bigan that trains the discriminator and encoder
        self.discriminator.trainable = False
 
        # Generate image from sampled noise
        z = Input(shape=(self.latent_dim, ))
        img_ = self.generator(z)
 
        # Encode image
        img = Input(shape=self.img_shape)
        z_ = self.encoder(img)
 
        # Latent -> img is fake, and img -> latent is valid
        fake = self.discriminator([z, img_])
        valid = self.discriminator([z_, img])
 
        # Set up and compile the combined model
        # Trains generator to fool the discriminator
        self.bigan_generator = Model([z, img], [fake, valid])
        self.bigan_generator.compile(loss=['binary_crossentropy', 'binary_crossentropy'],
            optimizer=optimizer)
 
 
    def build_encoder(self):
        model = Sequential()
 
        model.add(Flatten(input_shape=self.img_shape))
        model.add(Dense(512))
        model.add(LeakyReLU(alpha=0.2))
        model.add(BatchNormalization(momentum=0.8))
        model.add(Dense(512))
        model.add(LeakyReLU(alpha=0.2))
        model.add(BatchNormalization(momentum=0.8))
        model.add(Dense(self.latent_dim))
 
        model.summary()
 
        img = Input(shape=self.img_shape)
        z = model(img)
 
        return Model(img, z)
 
    def build_generator(self):
        model = Sequential()
 
        model.add(Dense(512, input_dim=self.latent_dim))
        model.add(LeakyReLU(alpha=0.2))
        model.add(BatchNormalization(momentum=0.8))
        model.add(Dense(512))
        model.add(LeakyReLU(alpha=0.2))
        model.add(BatchNormalization(momentum=0.8))
        model.add(Dense(np.prod(self.img_shape), activation='tanh'))
        model.add(Reshape(self.img_shape))
 
        model.summary()
 
        z = Input(shape=(self.latent_dim,))
        gen_img = model(z)
 
        return Model(z, gen_img)
 
    def build_discriminator(self):
 
        z = Input(shape=(self.latent_dim, ))
        img = Input(shape=self.img_shape)
        d_in = concatenate([z, Flatten()(img)])
 
        model = Dense(1024)(d_in)
        model = LeakyReLU(alpha=0.2)(model)
        model = Dropout(0.5)(model)
        model = Dense(1024)(model)
        model = LeakyReLU(alpha=0.2)(model)
        model = Dropout(0.5)(model)
        model = Dense(1024)(model)
        model = LeakyReLU(alpha=0.2)(model)
        model = Dropout(0.5)(model)
        validity = Dense(1, activation="sigmoid")(model)
 
        return Model([z, img], validity)
 
    def train(self, epochs, batch_size=128, sample_interval=50):
 
        # Load the dataset
        (X_train, _), (_, _) = mnist.load_data()
 
        # Rescale -1 to 1
        X_train = (X_train.astype(np.float32) - 127.5) / 127.5
        X_train = np.expand_dims(X_train, axis=3)
 
        # Adversarial ground truths
        valid = np.ones((batch_size, 1))
        fake = np.zeros((batch_size, 1))
 
        for epoch in range(epochs):
 
 
            # ---------------------
            #  Train Discriminator
            # ---------------------
 
            # Sample noise and generate img
            z = np.random.normal(size=(batch_size, self.latent_dim))
            imgs_ = self.generator.predict(z)
 
            # Select a random batch of images and encode
            idx = np.random.randint(0, X_train.shape[0], batch_size)
            imgs = X_train[idx]
            z_ = self.encoder.predict(imgs)
 
            # Train the discriminator (img -> z is valid, z -> img is fake)
            d_loss_real = self.discriminator.train_on_batch([z_, imgs], valid)
            d_loss_fake = self.discriminator.train_on_batch([z, imgs_], fake)
            d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
 
            # ---------------------
            #  Train Generator
            # ---------------------
 
            # Train the generator (z -> img is valid and img -> z is is invalid)
            g_loss = self.bigan_generator.train_on_batch([z, imgs], [valid, fake])
 
            # Plot the progress
            print ("%d [D loss: %f, acc: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss[0]))
 
            # If at save interval => save generated image samples
            if epoch % sample_interval == 0:
                self.sample_interval(epoch)
 
    def sample_interval(self, epoch):
        r, c = 5, 5
        z = np.random.normal(size=(25, self.latent_dim))
        gen_imgs = self.generator.predict(z)
 
        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("images/mnist_%d.png" % epoch)
        plt.close()