解决Keras GAN图像维度不匹配：生成器训练中的常见陷阱

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Reshape, Conv2DTranspose, Flatten
from tensorflow.keras.optimizers import Adam
import numpy as np
import cv2
import os

# 生成器模型
generator = Sequential()
generator.add(Dense(7 * 7 * 128, input_dim=100))
generator.add(Reshape((7, 7, 128)))
generator.add(Conv2DTranspose(64, kernel_size=5, strides=2, padding='same')) # 输出 14x14x64
generator.add(Conv2DTranspose(32, kernel_size=5, strides=2, padding='same')) # 输出 28x28x32
generator.add(Conv2DTranspose(3, kernel_size=5, activation='sigmoid', padding='same')) # 输出 28x28x3

import os
import cv2
import numpy as np

def load_images_color(directory, target_size=(224, 224)):
    """
    加载指定目录下的彩色图像，并调整大小、归一化。

    Args:
        directory (str): 图像文件所在的目录。
        target_size (tuple): 图像调整后的目标尺寸 (宽度, 高度)。

    Returns:
        np.array: 包含所有预处理后彩色图像的NumPy数组。
    """
    images = []
    for filename in os.listdir(directory):
        img_path = os.path.join(directory, filename)
        # 确保只处理图像文件，可以根据文件扩展名进行过滤
        if not (filename.lower().endswith(('.png', '.jpg', '.jpeg', '.bmp', '.gif'))):
            continue

        img = cv2.imread(img_path)
        if img is None: # 检查图像是否成功加载
            print(f"Warning: Could not load image {img_path}")
            continue

        img = cv2.resize(img, target_size)  # 调整图像尺寸
        img = img.astype('float32') / 255.0  # 归一化像素值到 [0, 1]
        images.append(img)
    return np.array(images)

# 假设彩色图像存储在 './ImageNet/dogs' 目录下
# 加载原始彩色图像，暂时调整为 224x224，以便后续灵活处理
cl_images = load_images_color('./ImageNet/dogs', target_size=(224, 224))
print(f"Loaded original color images shape: {cl_images.shape}") # 示例输出: (num_images, 224, 224, 3)

# 假设已经定义了 generator, optimizer, epochs, batch_size

# 编译生成器模型
optimizer = Adam(learning_rate=0.0002, beta_1=0.5)
generator.compile(loss='binary_crossentropy', optimizer=optimizer)

# 训练循环
epochs = 10000
batch_size = 32

for epoch in range(epochs):
    # 随机选择一个批次的索引
    idx = np.random.randint(0, cl_images.shape[0], batch_size)

    # 获取原始尺寸的彩色图像批次 (例如 224x224x3)
    cl_real_batch = cl_images[idx]

    # 将彩色图像批次调整为生成器输出的尺寸 (28x28x3)
    cl_real_small = []
    for im in cl_real_batch:
        cl_real_small.append(cv2.resize(im, (28, 28))) # 调整为 28x28
    cl_real_small = np.array(cl_real_small)

    # 检查调整后图像的形状
    # print(f"Shape of cl_real_small: {cl_real_small.shape}") # 预期输出: (32, 28, 28, 3)

    # 生成噪声输入
    noise = np.random.normal(0, 1, (batch_size, 100))

    # 训练生成器
    # 生成器现在将尝试生成与 cl_real_small 形状相同的图像
    g_loss = generator.train_on_batch(noise, cl_real_small)

    if epoch % 100 == 0:
        print(f"Epoch: {epoch}, Generator Loss: {g_loss}")