深度学习AAEGan学习笔记，keras版

我就是我 2022-05-13 18:47 278阅读 0赞

[对抗生成网络Gan变体集合 keras版本][Gan_ keras]

一.Adversarial Autoencoder GAN  
[https://arxiv.org/abs/1511.05644][https_arxiv.org_abs_1511.05644]  
aae的目的在于训练出自定义的编码器,自定义编码表现在real 真实样本的分布。  
如以下例子，目的就是在训练将一张图片进行编码，而该编码符合正太分布(np.random.normal)

aae组成  
1.编码器  
所谓编码器其实就是几层神经网络叠在一起，将一张(batch\_size,28,28)图片变成(batch\_size,10)向量  
2.解码器  
解码器就是将原来编码生成的(batch\_size,10)向量，解码还原成(batch\_size,28,28)图片  
3.符合正太分布(np.random.normal)+判别器D  
就是让编码器的编码规律符合正太分布

总结一句大白话:  
训练编码器，和解码器，而该编码器的编码方式符合正太分布。

代码位置:  
[https://github.com/eriklindernoren/Keras-GAN/tree/master/aae][https_github.com_eriklindernoren_Keras-GAN_tree_master_aae]

编码器Encoder(针对keras 版本做了些修改，需要自己写一个MergeLayer层)

class MergeLayer(Layer):
        
        def __init__(self, **kwargs):
    
            super(MergeLayer, self).__init__(**kwargs)
        def compute_output_shape(self, input_shape):
            
            return (input_shape[0][0], input_shape[0][1]) 
        def call(self, x, mask=None):
    
            final_output = x[0] + K.random_normal(K.shape(x[0])) * K.exp(x[1] / 2)
            return final_output
    
    def build_encoder(self):
            # Encoder
    
            img = Input(shape=self.img_shape)
    
            h = Flatten()(img)
            h = Dense(512)(h)
            h = LeakyReLU(alpha=0.2)(h)
            h = Dense(512)(h)
            h = LeakyReLU(alpha=0.2)(h)
            mu = Dense(self.latent_dim)(h)
            log_var = Dense(self.latent_dim)(h)
    # latent_repr = merge([mu, log_var],
    # mode=lambda p: p[0] + K.random_normal(K.shape(p[0])) * K.exp(p[1] / 2),
    # output_shape=lambda p: p[0])
    # output_shape=lambda p: p[0])
            latent_repr = MergeLayer()([mu,log_var])
            model = Model(img, latent_repr)
            plot_model(model, to_file='aae_encoder.png',show_shapes=True)
            return model

解码器Decoder

def build_decoder(self):
    
            model = Sequential()
    
            model.add(Dense(512, input_dim=self.latent_dim))
            model.add(LeakyReLU(alpha=0.2))
            model.add(Dense(512))
            model.add(LeakyReLU(alpha=0.2))
            model.add(Dense(np.prod(self.img_shape), activation='tanh'))
            model.add(Reshape(self.img_shape))
    
    
            z = Input(shape=(self.latent_dim,))
            img = model(z)
            return Model(z, img)

判别器D

def build_discriminator(self):
    
            model = Sequential()
    
            model.add(Dense(512, input_dim=self.latent_dim))
            model.add(LeakyReLU(alpha=0.2))
            model.add(Dense(256))
            model.add(LeakyReLU(alpha=0.2))
            model.add(Dense(1, activation="sigmoid"))
    
            encoded_repr = Input(shape=(self.latent_dim, ))
            validity = model(encoded_repr)
            return Model(encoded_repr, validity)

训练图:  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTEzMTEyOTE_size_27_color_FFFFFF_t_70]

[Gan_ keras]: https://blog.csdn.net/u011311291/article/details/82428620
[https_arxiv.org_abs_1511.05644]: https://arxiv.org/abs/1511.05644
[https_github.com_eriklindernoren_Keras-GAN_tree_master_aae]: https://github.com/eriklindernoren/Keras-GAN/tree/master/aae
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTEzMTEyOTE_size_27_color_FFFFFF_t_70]: /images/20220504/3589c4384b9741e1b741088837c44772.png