7. 使用CycleGAN进行图像转换

槿灵兮

于 2024-08-27 04:15:39 发布

阅读量22

点赞数

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.layers import Conv2D, Conv2DTranspose, LeakyReLU, BatchNormalization, Input, Concatenate
from tensorflow.keras.models import Model

def build_generator():
    inputs = Input(shape=(256, 256, 3))
    x = Conv2D(64, 4, strides=2, padding='same', activation='relu')(inputs)
    x = BatchNormalization()(x)
    x = Conv2D(128, 4, strides=2, padding='same', activation='relu')(x)
    x = BatchNormalization()(x)
    x = Conv2D(256, 4, strides=2, padding='same', activation='relu')(x)
    x = BatchNormalization()(x)
    x = Conv2D(512, 4, strides=2, padding='same', activation='relu')(x)
    x = BatchNormalization()(x)
    x = Conv2DTranspose(256, 4, strides=2, padding='same', activation='relu')(x)
    x = BatchNormalization()(x)
    x = Conv2DTranspose(128, 4, strides=2, padding='same', activation='relu')(x)
    x = BatchNormalization()(x)
    x = Conv2DTranspose(64, 4, strides=2, padding='same', activation='relu')(x)
    x = BatchNormalization()(x)
    outputs = Conv2D(3, 4, strides=1, padding='same', activation='tanh')(x)
    return Model(inputs, outputs)

def build_discriminator():
    inputs = Input(shape=(256, 256, 3))
    x = Conv2D(64, 4, strides=2, padding='same', activation='relu')(inputs)
    x = LeakyReLU(alpha=0.2)(x)
    x = Conv2D(128, 4, strides=2, padding='same', activation='relu')(x)
    x = LeakyReLU(alpha=0.2)(x)
    x = Conv2D(256, 4, strides=2, padding='same', activation='relu')(x)
    x = LeakyReLU(alpha=0.2)(x)
    x = Conv2D(512, 4, strides=2, padding='same', activation='relu')(x)
    x = LeakyReLU(alpha=0.2)(x)
    x = Conv2D(1, 4, strides=1, padding='same')(x)
    return Model(inputs, x)

# Hyperparameters
epochs = 20
batch_size = 1

# Load dataset (example)
def load_data():
    # Placeholder function to load dataset
    return np.random.rand(10, 256, 256, 3), np.random.rand(10, 256, 256, 3)

# Initialize models
generator_g = build_generator()
generator_f = build_generator()
discriminator_x = build_discriminator()
discriminator_y = build_discriminator()

# Compile models
generator_g.compile(optimizer=tf.keras.optimizers.Adam(1e-4))
generator_f.compile(optimizer=tf.keras.optimizers.Adam(1e-4))
discriminator_x.compile(optimizer=tf.keras.optimizers.Adam(1e-4))
discriminator_y.compile(optimizer=tf.keras.optimizers.Adam(1e-4))

# Training loop
for epoch in range(epochs):
    real_images_x, real_images_y = load_data()
    fake_images_y = generator_g.predict(real_images_x)
    fake_images_x = generator_f.predict(real_images_y)
    
    # Train Discriminators
    d_loss_real_x = discriminator_x.train_on_batch(real_images_x, np.ones((batch_size, 256, 256, 1)))
    d_loss_fake_x = discriminator_x.train_on_batch(fake_images_x, np.zeros((batch_size, 256, 256, 1)))
    
    d_loss_real_y = discriminator_y.train_on_batch(real_images_y, np.ones((batch_size, 256, 256, 1)))
    d_loss_fake_y = discriminator_y.train_on_batch(fake_images_y, np.zeros((batch_size, 256, 256, 1)))
    
    # Train Generators
    g_loss_g = generator_g.train_on_batch(real_images_x, np.ones((batch_size, 256, 256, 1)))
    g_loss_f = generator_f.train_on_batch(real_images_y, np.ones((batch_size, 256, 256, 1)))
    
    print(f'Epoch [{epoch+1}/{epochs}], D Loss X Real: {d_loss_real_x}, D Loss X Fake: {d_loss_fake_x}, D Loss Y Real: {d_loss_real_y}, D Loss Y Fake: {d_loss_fake_y}')
    print(f'G Loss G: {g_loss_g}, G Loss F: {g_loss_f}')
    
    if (epoch + 1) % 5 == 0:
        output_images = generator_g.predict(real_images_x)
        for i in range(batch_size):
            plt.imshow(output_images[i])
            plt.axis('off')
            plt.savefig(f'cyclegan_image_{epoch+1}_{i}.png')
            plt.close()