深度学习ACGan学习笔记，keras版

痛定思痛。 2022-05-14 08:55 327阅读 0赞

[对抗生成网络Gan变体集合 keras版本][Gan_ keras]

一.ACGAN(Auxiliary Classifier GAN)  
[https://arxiv.org/abs/1610.09585][https_arxiv.org_abs_1610.09585]  
依旧有Generator,Discriminator,可使用MNSIT训练生成图片。

和DCGAN的不同:  
1.增加了class类别标签参与训练，可以生成指定类别的图片

代码引用的《Web安全之强化学习与GAN》，位置:  
[https://github.com/duoergun0729/3book/tree/master/code/keras-acgan.py][https_github.com_duoergun0729_3book_tree_master_code_keras-acgan.py]

生成器G代码:

def build_generator(latent_size):
    
        cnn = Sequential()
        cnn.add(Dense(1024, input_dim=latent_size, activation='relu'))
        cnn.add(Dense(128 * 7 * 7, activation='relu'))
        cnn.add(Reshape((128, 7, 7)))
        cnn.add(UpSampling2D(size=(2, 2)))
        cnn.add(Conv2D(256, (5, 5), padding="same", kernel_initializer="glorot_normal", activation="relu"))
        cnn.add(UpSampling2D(size=(2, 2)))
        cnn.add(Conv2D(128, (5, 5), padding="same", kernel_initializer="glorot_normal", activation="relu"))
        cnn.add(Conv2D(1, (2, 2), padding="same", kernel_initializer="glorot_normal", activation="tanh"))
        latent = Input(shape=(latent_size, ))
        image_class = Input(shape=(1,), dtype='int32')
        cls = Flatten()(Embedding(10, 100, embeddings_initializer="glorot_normal")(image_class))
        #h = merge([latent, cls], mode='mul')
        h=add([latent, cls])
        fake_image = cnn(h)
        return Model(inputs=[latent, image_class], outputs=[fake_image])

判别器D代码:

def build_discriminator():
    
        cnn = Sequential()
        cnn.add(Conv2D(32, (3, 3), padding="same", strides=(2, 2), input_shape=(1, 28, 28) ))
        cnn.add(LeakyReLU())
        cnn.add(Dropout(0.3))
        cnn.add(Conv2D(64, (3, 3), padding="same", strides=(1, 1)))
        cnn.add(LeakyReLU())
        cnn.add(Dropout(0.3))
        cnn.add(Conv2D(128, (3, 3), padding="same", strides=(2, 2)))
        cnn.add(LeakyReLU())
        cnn.add(Dropout(0.3))
        cnn.add(Conv2D(256, (3, 3), padding="same", strides=(1, 1)))
        cnn.add(LeakyReLU())
        cnn.add(Dropout(0.3))
        cnn.add(Flatten())
        image = Input(shape=(1, 28, 28))
        features = cnn(image)
        fake = Dense(1, activation='sigmoid', name='generation')(features)
        aux = Dense(10, activation='softmax', name='auxiliary')(features)
        return Model(inputs=[image], outputs=[fake, aux])

训练图:  
![这里写图片描述][70]

[Gan_ keras]: https://blog.csdn.net/u011311291/article/details/82428620
[https_arxiv.org_abs_1610.09585]: https://arxiv.org/abs/1610.09585
[https_github.com_duoergun0729_3book_tree_master_code_keras-acgan.py]: https://github.com/duoergun0729/3book/tree/master/code/keras-acgan.py
[70]: /images/20220514/6fe263b1365547b2b9763d1bd3d6d012.png