CGAN

# 同原始GAN
#  生成器，与判别器改为使用CNN网络

import numpy as np
import matplotlib
from matplotlib import pyplot as plt
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers,optimizers,losses
from tensorflow.keras.callbacks import EarlyStopping
from IPython.display import Image
import cv2
import PIL
import json, os

# 定义全局变量
img_shape = (28, 28, 1)
batch_size = 200
latent_dim = 100
sample_interval=1
epochs=10

# 定义生成器
def build_generator():

        input=keras.Input(shape=latent_dim)

        x=layers.Dense(256*7*7)(input)
        x=layers.BatchNormalization(momentum=0.8)(x)
        x=layers.Activation(activation='relu')(x)
        x=layers.Reshape((7,7,256))(x)

        x=layers.UpSampling2D((2,2))(x)
        x = layers.Conv2D(128, (3, 3), padding='same')(x)
        x = layers.Conv2D(128, (3, 3), padding='same')(x)
        x = layers.BatchNormalization(momentum=0.8)(x)
        x = layers.Activation(activation='relu')(x)

        x = layers.UpSampling2D((2, 2))(x)
        x = layers.Conv2D(64, (3, 3), padding='same')(x)
        x = layers.Conv2D(64, (3, 3), padding='same')(x)
        x = layers.BatchNormalization(momentum=0.8)(x)
        x = layers.Activation(activation='relu')(x)

        x = layers.Conv2D(32, (3, 3), padding='same')(x)
        x = layers.Conv2D(32, (3, 3), padding='same')(x)
        x = layers.BatchNormalization(momentum=0.8)(x)
        x = layers.Activation(activation='relu')(x)

        output = layers.Conv2D(1, (3, 3), padding='same',activation='sigmoid')(x)

        model=keras.Model(inputs=input,outputs=output,name='generator')
        model.summary()
        return model

def build_discriminator():#判别器
        input=keras.Input(shape=img_shape)
        x = layers.Conv2D(32, (3, 3), padding='same')(input)
        x = layers.Conv2D(32, (3, 3), padding='same')(x)
        x = layers.LeakyReLU(alpha=0.2)(x)
        x = layers.Dropout(0.25)(x)
        x = layers.MaxPooling2D(2,2)(x)

        x = layers.Conv2D(64, (3, 3), padding='same')(x)
        x = layers.Conv2D(64, (3, 3), padding='same')(x)
        x = layers.LeakyReLU(alpha=0.2)(x)
        x = layers.Dropout(0.25)(x)
        x = layers.MaxPooling2D(2, 2)(x)

        x = layers.Conv2D(128, (3, 3), padding='same')(x)
        x = layers.Conv2D(128, (3, 3), padding='same')(x)
        x = layers.LeakyReLU(alpha=0.2)(x)
        x = layers.Dropout(0.25)(x)
        x = layers.MaxPooling2D(2, 2)(x)

        x = layers.Conv2D(256, (3, 3), padding='same')(x)
        x = layers.Conv2D(256, (3, 3), padding='same')(x)
        x = layers.LeakyReLU(alpha=0.2)(x)
        x = layers.Dropout(0.25)(x)

        x=layers.Flatten()(x)
        output=layers.Dense(1,activation='sigmoid')(x)

        model=keras.Model(inputs=input,outputs=output,name='discriminator')
        model.summary()
        return model

generator_model = build_generator()
discriminator_model = build_discriminator()

Model: "generator"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_4 (InputLayer)         [(None, 100)]             0         
_________________________________________________________________
dense_2 (Dense)              (None, 12544)             1266944   
_________________________________________________________________
batch_normalization_4 (Batch (None, 12544)             50176     
_________________________________________________________________
activation_4 (Activation)    (None, 12544)             0         
_________________________________________________________________
reshape_1 (Reshape)          (None, 7, 7, 256)         0         
_________________________________________________________________
up_sampling2d_2 (UpSampling2 (None, 14, 14, 256)       0         
_________________________________________________________________
conv2d_15 (Conv2D)           (None, 14, 14, 128)       295040    
_________________________________________________________________
conv2d_16 (Conv2D)           (None, 14, 14, 128)       147584    
_________________________________________________________________
batch_normalization_5 (Batch (None, 14, 14, 128)       512       
_________________________________________________________________
activation_5 (Activation)    (None, 14, 14, 128)       0         
_________________________________________________________________
up_sampling2d_3 (UpSampling2 (None, 28, 28, 128)       0         
_________________________________________________________________
conv2d_17 (Conv2D)           (None, 28, 28, 64)        73792     
_________________________________________________________________
conv2d_18 (Conv2D)           (None, 28, 28, 64)        36928     
_________________________________________________________________
batch_normalization_6 (Batch (None, 28, 28, 64)        256       
_________________________________________________________________
activation_6 (Activation)    (None, 28, 28, 64)        0         
_________________________________________________________________
conv2d_19 (Conv2D)           (None, 28, 28, 32)        18464     
_________________________________________________________________
conv2d_20 (Conv2D)           (None, 28, 28, 32)        9248      
_________________________________________________________________
batch_normalization_7 (Batch (None, 28, 28, 32)        128       
_________________________________________________________________
activation_7 (Activation)    (None, 28, 28, 32)        0         
_________________________________________________________________
conv2d_21 (Conv2D)           (None, 28, 28, 1)         289       
=================================================================
Total params: 1,899,361
Trainable params: 1,873,825
Non-trainable params: 25,536
_________________________________________________________________
Model: "discriminator"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_5 (InputLayer)         [(None, 28, 28, 1)]       0         
_________________________________________________________________
conv2d_22 (Conv2D)           (None, 28, 28, 32)        320       
_________________________________________________________________
conv2d_23 (Conv2D)           (None, 28, 28, 32)        9248      
_________________________________________________________________
leaky_re_lu_4 (LeakyReLU)    (None, 28, 28, 32)        0         
_________________________________________________________________
dropout_4 (Dropout)          (None, 28, 28, 32)        0         
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 14, 14, 32)        0         
_________________________________________________________________
conv2d_24 (Conv2D)           (None, 14, 14, 64)        18496     
_________________________________________________________________
conv2d_25 (Conv2D)           (None, 14, 14, 64)        36928     
_________________________________________________________________
leaky_re_lu_5 (LeakyReLU)    (None, 14, 14, 64)        0         
_________________________________________________________________
dropout_5 (Dropout)          (None, 14, 14, 64)        0         
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 7, 7, 64)          0         
_________________________________________________________________
conv2d_26 (Conv2D)           (None, 7, 7, 128)         73856     
_________________________________________________________________
conv2d_27 (Conv2D)           (None, 7, 7, 128)         147584    
_________________________________________________________________
leaky_re_lu_6 (LeakyReLU)    (None, 7, 7, 128)         0         
_________________________________________________________________
dropout_6 (Dropout)          (None, 7, 7, 128)         0         
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 3, 3, 128)         0         
_________________________________________________________________
conv2d_28 (Conv2D)           (None, 3, 3, 256)         295168    
_________________________________________________________________
conv2d_29 (Conv2D)           (None, 3, 3, 256)         590080    
_________________________________________________________________
leaky_re_lu_7 (LeakyReLU)    (None, 3, 3, 256)         0         
_________________________________________________________________
dropout_7 (Dropout)          (None, 3, 3, 256)         0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 2304)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 1)                 2305      
=================================================================
Total params: 1,173,985
Trainable params: 1,173,985
Non-trainable params: 0
_________________________________________________________________

# 建立GAN模型
def bulid_model():
      discriminator_model.compile(loss='binary_crossentropy',
                    optimizer=keras.optimizers.Adam(0.0001, 0.000001),
                    metrics=['accuracy'])
      
      #使判别器不训练，固定判别器
      discriminator_model.trainable = False
      inputs = keras.Input(shape=latent_dim)
      img = generator_model(inputs)
      outputs = discriminator_model(img)
      model = keras.Model(inputs=inputs, outputs=outputs)
      model.summary()
      # 训练GAN
      model.compile(optimizer=keras.optimizers.Adam(0.0001, 0.000001),
                    loss='binary_crossentropy',
                    )
      return model

# 模型实例化
model =bulid_model()

Model: "model_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_6 (InputLayer)         [(None, 100)]             0         
_________________________________________________________________
generator (Functional)       (None, 28, 28, 1)         1899361   
_________________________________________________________________
discriminator (Functional)   (None, 1)                 1173985   
=================================================================
Total params: 3,073,346
Trainable params: 1,873,825
Non-trainable params: 1,199,521
_________________________________________________________________

# 数据加载
def load_data():
        (train_images, train_labels), (test_images, test_labels) = keras.datasets.mnist.load_data()
        train_images = train_images /255
        train_images = np.expand_dims(train_images, axis=3)
        print('img_number:',train_images.shape)
        return train_images
 #生成图片
def generate_sample_images(epoch):  

      row, col = 5, 5#
      noise = np.random.normal(0, 1, (row * col, 100))
      #生成噪声
      gan_imgs = generator_model.predict(noise)
      fig, axs = plt.subplots(row, col)
      #生成5*5的画板
      idx = 0
      for i in range(row):
          for j in range(col):
              axs[i, j].imshow(gan_imgs[idx, :, :, 0], cmap='gray')
              axs[i, j].axis('off')
              idx += 1
      plt.show()

# 定义训练
def train():
    train_images=load_data()

    #生成标签
    valid = np.ones((batch_size, 1))
    fake = np.zeros((batch_size, 1))

    step=int(train_images.shape[0]/batch_size)
    #计算步长
    print('step:',step)

    for epoch in range(epochs):
        print(epoch)
        train_images = (tf.random.shuffle(train_images)).numpy()
        #每个epoch打乱一次
        if epoch % 2 == 0:
          # 图片展示
            generate_sample_images(epoch)

        for i in range(step):

            idx = np.arange(i*batch_size,i*batch_size+batch_size,1)
            #生成索引
            imgs =train_images[idx]
            #读取索引对应的图片
            noise = np.random.normal(0, 1, (batch_size, 100))
            gan_imgs = generator_model.predict(noise)
            #通过噪声生成图片
            #训练判别器
            discriminator_loss_real = discriminator_model.train_on_batch(imgs, valid)  # 真实数据对应标签1
            discriminator_loss_fake = discriminator_model.train_on_batch(gan_imgs, fake)  # 生成的数据对应标签0
            discriminator_loss = 0.5 * np.add(discriminator_loss_real, discriminator_loss_fake)
            #训练生成器
            noise = np.random.normal(0, 1, (batch_size, 100))
            # noise = np.random.normal(0, 1, (1000, 100))
            # 判别器训练一回，生成器生成5回
            # for j in range(5):
            #   a=j*200
            #   b=(j+1)*200
            #   noise_temp=noise[a:b]
            generator_loss = model.train_on_batch(noise, valid)
            if i%100==0:#每100步进行输出
                print("epoch:%d step:%d [discriminator_loss: %f, acc: %.2f%%] [generator_loss: %f]" % (
                    epoch,i,discriminator_loss[0], 100 * discriminator_loss[1], generator_loss))

if __name__ == '__main__':
   train()

img_number: (60000, 28, 28, 1)
step: 300
0

epoch:0 step:0 [discriminator_loss: 0.686799, acc: 49.25%] [generator_loss: 0.717094]
epoch:0 step:100 [discriminator_loss: 0.488994, acc: 78.00%] [generator_loss: 2.385789]
epoch:0 step:200 [discriminator_loss: 0.671490, acc: 57.25%] [generator_loss: 1.069866]
1
epoch:1 step:0 [discriminator_loss: 0.684170, acc: 55.00%] [generator_loss: 0.862875]
epoch:1 step:100 [discriminator_loss: 0.689809, acc: 50.75%] [generator_loss: 0.832214]
epoch:1 step:200 [discriminator_loss: 0.680106, acc: 58.50%] [generator_loss: 0.805226]
2

epoch:2 step:0 [discriminator_loss: 0.668721, acc: 63.25%] [generator_loss: 0.823960]
epoch:2 step:100 [discriminator_loss: 0.680783, acc: 55.25%] [generator_loss: 0.829456]
epoch:2 step:200 [discriminator_loss: 0.672865, acc: 59.00%] [generator_loss: 0.825836]
3
epoch:3 step:0 [discriminator_loss: 0.640733, acc: 63.25%] [generator_loss: 0.906384]
epoch:3 step:100 [discriminator_loss: 0.658660, acc: 57.00%] [generator_loss: 0.879481]
epoch:3 step:200 [discriminator_loss: 0.635054, acc: 62.75%] [generator_loss: 0.930342]
4

epoch:4 step:0 [discriminator_loss: 0.643901, acc: 62.75%] [generator_loss: 0.902657]
epoch:4 step:100 [discriminator_loss: 0.628503, acc: 63.00%] [generator_loss: 0.956084]
epoch:4 step:200 [discriminator_loss: 0.622460, acc: 65.25%] [generator_loss: 1.016035]
5
epoch:5 step:0 [discriminator_loss: 0.665893, acc: 60.75%] [generator_loss: 0.991473]
epoch:5 step:100 [discriminator_loss: 0.632177, acc: 63.75%] [generator_loss: 1.019901]
epoch:5 step:200 [discriminator_loss: 0.622525, acc: 66.00%] [generator_loss: 0.973664]