ImageDataGenerator无法检测图像问题咨询：路径正确却无识别

我尝试对dirty、vehicle、clean三类图像做数据增强，计划将增强后的图像保存至同结构的目标文件夹，且已搭建好文件夹结构，但运行代码时无法检测到图像。

原代码 #

import os
from tensorflow.keras.preprocessing.image import ImageDataGenerator

# Set the paths to your dataset folders
data_dir = "\/home\/intern\/Desktop\/automated-py-scripts\/real_data"
new_augment_dir = "\/home\/intern\/Desktop\/automated-py-scripts\/augmented"
train_dir = os.path.join(data_dir, "train")

# Define the augmentation parameters
augmentation_params = {
    "rotation_range": 15,
    "width_shift_range": 0.1,
    "height_shift_range": 0.1,
    "shear_range": 0.1,
    # "zoom_range": 0.1,
    "horizontal_flip": True,
    "fill_mode": "nearest"
}

# Create an ImageDataGenerator instance with augmentation parameters
datagen = ImageDataGenerator(**augmentation_params)

# Get the list of class folders
class_folders = ['clean', 'dirty', 'vehicle']

# Augment images in each class folder
for folder in class_folders:
    class_dir = os.path.join(train_dir, folder)
    print(class_dir)
    save_dir = os.path.join(new_augment_dir, folder)

    # Create the save directory if it doesn't exist
    os.makedirs(save_dir, exist_ok=True)

    # Create a generator for the images in the class folder
    generator = datagen.flow_from_directory(
        class_dir,
        target_size=(224, 224),  # Resize the images to your desired size
        batch_size=1,
        class_mode=None,
        save_to_dir=save_dir,
        save_format="jpg",
        shuffle=True
    )

    # Generate augmented images and save them to the save directory
    num_images = len(os.listdir(class_dir))
    num_augmented_images = 3  # Number of augmented images to generate per original image

    for _ in range(num_images * num_augmented_images):
        images = next(generator)

运行错误输出 #

/home/intern/Desktop/automated-py-scripts/real_data/train/clean
Found 0 images belonging to 0 classes.
/home/intern/Desktop/automated-py-scripts/real_data/train/dirty
Found 0 images belonging to 0 classes.
/home/intern/Desktop/automated-py-scripts/real_data/train/vehicle
Found 0 images belonging to 0 classes.

问题原因 #

flow_from_directory的核心要求是：输入目录下必须嵌套一层类别子目录，哪怕是单类别场景。你的当前目录结构是train/clean/直接存放图片，没有额外的子目录，导致函数无法识别类别和图像。

解决方案 #

方案一：调整文件夹结构（推荐，符合Keras标准格式） #

给每个类别文件夹新增一层同名子目录，将原图片放入该子目录中，最终结构如下：

real_data/train/
    clean/
        clean/  # 新增子目录
            img1.jpg
            img2.jpg
            ...
    dirty/
        dirty/
            img1.jpg
            ...
    vehicle/
        vehicle/
            img1.jpg
            ...

调整后，原代码无需修改即可正常识别图像。

方案二：修改代码，绕开嵌套目录要求 #

如果不想调整文件夹结构，可以直接遍历每个类别下的图片文件，手动读取后用datagen.flow()生成增强图，修改后的代码如下：

import os
from tensorflow.keras.preprocessing.image import ImageDataGenerator, load_img, img_to_array

data_dir = "/home/intern/Desktop/automated-py-scripts/real_data"
new_augment_dir = "/home/intern/Desktop/automated-py-scripts/augmented"
train_dir = os.path.join(data_dir, "train")

augmentation_params = {
    "rotation_range": 15,
    "width_shift_range": 0.1,
    "height_shift_range": 0.1,
    "shear_range": 0.1,
    "horizontal_flip": True,
    "fill_mode": "nearest"
}

datagen = ImageDataGenerator(**augmentation_params)
class_folders = ['clean', 'dirty', 'vehicle']

for folder in class_folders:
    class_dir = os.path.join(train_dir, folder)
    print(class_dir)
    save_dir = os.path.join(new_augment_dir, folder)
    os.makedirs(save_dir, exist_ok=True)

    # 筛选出当前目录下的图像文件（支持jpg/jpeg/png）
    img_files = [f for f in os.listdir(class_dir) if f.lower().endswith(('.jpg', '.jpeg', '.png'))]
    num_augmented_images = 3  # 每张原图生成的增强图数量

    for img_file in img_files:
        img_path = os.path.join(class_dir, img_file)
        # 加载并预处理图像
        img = load_img(img_path, target_size=(224, 224))
        x = img_to_array(img)
        x = x.reshape((1,) + x.shape)  # 转换为(batch_size, height, width, channels)格式

        # 生成增强图像并保存
        i = 0
        for batch in datagen.flow(x, batch_size=1, save_to_dir=save_dir, save_prefix='aug_', save_format='jpg'):
            i += 1
            if i >= num_augmented_images:
                break  # 达到指定数量后停止生成

方案对比 #

• 方案一：符合Keras官方推荐的数据格式，后续如果用flow_from_directory做模型训练会更顺畅。
• 方案二：无需修改现有文件夹结构，代码更灵活，适合快速验证场景。

内容的提问来源于stack exchange，提问作者NevMthw