DCGAN生成动漫头像【学习】
DCGAN生成动漫头像
在假期看了李宏毅老师的GAN的介绍，看到了课后题DCGAN生成动漫头像的作业，实现一下。记录学习过程。
参考的文章：
【Keras】 基于GAN自动生成动漫头像
因为使用的是tensorflow-gpu，运行过程中发现GPU显存占满电脑特别卡的问题，参考这篇文章（tensorflow使用GPU训练时的显存占用问题）加了几行代码，解决了GPU显存占满卡死的问题。
具体所加代码如下：
from tensorflow.compat.v1 import ConfigProto
from tensorflow.compat.v1 import InteractiveSession
config = ConfigProto()
config.gpu_options.allow_growth = True
session = InteractiveSession(config=config)
这样显存就不会占满，不容易卡死，跑程序的时候也可以在电脑上做别的事情。
数据集的话尝试了一下用其他博主的爬虫去爬动漫网站的图，发现很慢，于是直接下载了现有的数据集,这里的图都是裁剪好的96*96的，可以直接拿来训练，附上网盘链接：
链接: https://pan.baidu.com/s/1zLVnmH3Qw56S7jBzlIIeCg 提取码: if8g
先附上代码：
from keras.datasets import mnist
from keras.layers import Input,Dense,Reshape,Flatten,Dropout
from keras.layers import BatchNormalization,Activation,ZeroPadding2D
from keras.layers import LeakyReLU
from keras.layers import UpSampling2D,Conv2D
from keras.models import Sequential,Model
from keras.optimizers import Adam,RMSprop
from keras.models import load_model
from keras.preprocessing import image
import os
import matplotlib.pyplot as plt
import sys
import numpy as np
from tensorflow.compat.v1 import ConfigProto
from tensorflow.compat.v1 import InteractiveSession
#设定GPU按需分配内存，不让GPU内存占满导致死机
config = ConfigProto()
config.gpu_options.allow_growth = True
session = InteractiveSession(config=config)
class DCGAN():
def __init__(self):
#训练集图像的数据
self.img_rows = 96;
self.img_cols = 96;
self.channels = 3;
self.img_shape=(self.img_rows,self.img_cols,self.channels)
self.latent_dim =100
optimizer = Adam(0.0002,0.5)
optimizerD =RMSprop(lr=0.0008, clipvalue=1.0, decay=6e-8)
optimizerG = RMSprop(lr=0.0004, clipvalue=1.0, decay=6e-8)
print("-1")
#对判别器进行构建和编译
self.discriminator = self.build_discriminator()
print("0")
self.discriminator.compile(loss='binary_crossentropy',optimizer=optimizer,metrics=['accuracy'])
print("1")
#对生成器进行构造
self.generator = self.build_generator()
print("2")
# The generator takes noise as input and generates imgs
z = Input(shape=(self.latent_dim,))
img = self.generator(z)
# 总体模型只对生成器进行训练
self.discriminator.trainable = False
# 从生成器中生成的图 经过判别器获得一个valid
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(512*6*6,activation='relu',input_dim=self.latent_dim)) #输入维度为100，输出128*7*7
model.add(Reshape((6,6,512)))
model.add(UpSampling2D()) #进行上采样，变成14*14*128
model.add(Conv2D(256,kernel_size=5,padding='same'))
model.add(BatchNormalization(momentum=0.8))#该层在每个batch上将前一层的激活值重新规范化，即使得其输出数据的均值接近0，其标准差接近1优点（1）加速收敛 （2）控制过拟合，可以少用或不用Dropout和正则 （3）降低网络对初始化权重不敏感 （4）允许使用较大的学习率
model.add(Activation("relu"))#
model.add(UpSampling2D())
model.add(Conv2D(128, kernel_size=5, padding="same"))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation("relu"))
model.add(UpSampling2D())
model.add(Conv2D(64, kernel_size=5, padding="same"))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation("relu"))
model.add(UpSampling2D())
model.add(Conv2D(self.channels, kernel_size=5, padding="same"))
model.add(Activation("tanh"))
model.summary() #打印网络参数
noise = Input(shape=(self.latent_dim,))
img = model(noise)
return Model(noise,img) #定义一个 一个输入noise一个输出img的模型
def build_discriminator(self):
dropout = 0.25
depth = 32
model = Sequential()
# model.add(Conv2D(depth,kernel_size=3,strides=2,input_shape=self.img_shape, padding="same"))
# model.add(LeakyReLU(alpha=0.2)) #alpha：大于0的浮点数，代表激活函数图像中第三象限线段的斜率
# model.add(Dropout(dropout))
# model.add(Conv2D(depth*2, kernel_size=3, strides=2, padding="same"))
# model.add(LeakyReLU(alpha=0.2))
# model.add(Dropout(dropout))
# model.add(Conv2D(depth * 4, kernel_size=3, strides=2, padding="same"))
# model.add(LeakyReLU(alpha=0.2))
# model.add(Dropout(dropout))
# model.add(Conv2D(depth * 8, kernel_size=3, strides=1, padding="same"))
# model.add(LeakyReLU(alpha=0.2))
# model.add(Dropout(dropout))
# model.add(Flatten())
# model.add(Dense(1,activation='sigmoid'))
model.add(Conv2D(64, kernel_size=5, strides=2, input_shape=self.img_shape, padding="same"))
model.add(LeakyReLU(alpha=0.2))
model.add(Dropout(0.25))
model.add(Conv2D(128, kernel_size=5, 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(256, kernel_size=5, strides=2, padding="same"))
model.add(BatchNormalization(momentum=0.8))
model.add(LeakyReLU(alpha=0.2))
model.add(Dropout(0.25))
model.add(Conv2D(512, kernel_size=5, 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,epochs,batch_size=128,save_interval = 50):
# Adversarial ground truths
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
for epoch in range(epochs):
# ---------------------
# Train Discriminator
# ---------------------
# Select a random half of images
# idx = np.random.randint(0, X_train.shape[0], batch_size)
# imgs = X_train[idx]
imgs = self.load_batch_imgs(batch_size,'faces')
# Sample noise and generate a batch of new images
noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
gen_imgs = self.generator.predict(noise)
# Train the discriminator (real classified as ones and generated as zeros)
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)
# ---------------------
# Train Generator
# ---------------------
# Train the generator (wants discriminator to mistake images as real)
g_loss = self.combined.train_on_batch(noise, valid)
# Plot the progress
print ("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss))
# If at save interval => save generated image samples
if epoch % save_interval == 0:
self.combined.save('./model/combined_model_%d.h5'%epoch)
self.discriminator.save('./model/discriminator_model_%d.h5'%epoch )
self.save_imgs(epoch)
def load_batch_imgs(self,batch_size,dirName):
img_names = os.listdir(os.path.join(dirName))
img_names = np.array(img_names)
idx = np.random.randint(0, img_names.shape[0], batch_size)
img_names = img_names[idx]
img = []
# 把图片读取出来放到列表中
for i in range(len(img_names)):
images = image.load_img(os.path.join(dirName, img_names[i]), target_size=(96, 96))
x = image.img_to_array(images)
x = np.expand_dims(x, axis=0)
img.append(x)
# print('loading no.%s image' % i)
# 把图片数组联合在一起
x = np.concatenate([x for x in img])
x = x / 127.5 - 1.
return x
def save_imgs(self, epoch):
r, c = 5, 5
noise = np.random.normal(0, 1, (r * c, self.latent_dim)) #高斯分布，均值0，标准差1，size= (5*5, 100)
gen_imgs = self.generator.predict(noise)
# Rescale images 0 - 1
gen_imgs = 0.5 * gen_imgs + 0.5
fig, axs = plt.subplots(r, c)
cnt = 0 #生成的25张图 显示出来
for i in range(r):
for j in range(c):
axs[i,j].imshow(gen_imgs[cnt, :,:,:])
axs[i,j].axis('off')
cnt += 1
fig.savefig("images/mnist_%d.png" % epoch)
plt.close()
def loadModel(self):
self.combined = load_model('./model/combined_model_last.h5')
self.discriminator = load_model('./model/discriminator_model_last.h5')
if __name__ == '__main__':
dcgan = DCGAN()
dcgan.train(epochs=40000, batch_size=64, save_interval=200)
最后一行的epochs应该是训练GAN网络的次数，可以根据需求修改，但实际过程中我发现到了20000效果就比较好了，多了可能会过拟合导致训练效果变差。batch_size大的话可以加快训练速度，但是可能会导致内存不足，太小的话可能会导致训练变慢，参考batch size设置技巧 谈谈batchsize参数，第三个参数save_interval顾名思义就是保存间隔，每隔200个epoch保存一次模型和图片，大小可以自己设定，200有点小，可以500。
训练结果
这是大约epoch=22000时的输出
结果还比较合理，代码还需要详细阅读。