从图像本地目录创建tensorflow数据集

import tensorflow as tf
import os

image_dir = 'path/to/image/directory'

image_paths = [os.path.join(image_dir, filename) for filename in os.listdir(image_dir)]

labels = ['label1', 'label2', ...]  # 标签列表

def parse_image(filename):
    image_string = tf.io.read_file(filename)
    image = tf.image.decode_image(image_string, channels=3)  # 根据图像类型进行解码
    # 进行图像预处理操作（如缩放、裁剪、归一化等）
    return image

def create_dataset(image_paths, labels=None):
    dataset = tf.data.Dataset.from_tensor_slices(image_paths)
    dataset = dataset.map(parse_image, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    if labels is not None:
        dataset = dataset.map(lambda x: (x, tf.constant(labels)))  # 添加标签
    return dataset

dataset = create_dataset(image_paths, labels)
# 进行数据集的批处理、乱序、重复等操作
# 进行模型训练或其他任务