（Python）扩充样本图片及其label的数量

通过调整亮度、大小等，分别扩充样本和对应label的数量

（需安装torchvision）

import torchvision.transforms as transforms
from PIL import Image
import torchvision

pic_read = '样本文件地址'
label_read = '对应label地址'
pic_save = '样本扩充保存地址'
label_save = 'label扩充保存地址'

# def voc_rand_crop(feature, label, height, wide):
#     rect = torchvision.transforms.RandomCrop.get_params(feature, (height, wide))
#     feature = torchvision.transforms.functional.crop(feature, *rect)
#     # print(type(feature))
#     label = torchvision.transforms.functional.crop(label, *rect)
#     return feature, label




def main():

    transform = transforms.ColorJitter(brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5)
    transform2 = transforms.Grayscale(3)

    for i in range(1):
        img = Image.open(pic_read + f'{i}'+'.png')
        img2 = Image.open(label_read + f'{i}'+'.png')


        for i in range(5000):#扩充的数量
            imgt = transform(img)
            imgt2 = transform2(imgt)
            rect = torchvision.transforms.RandomCrop.get_params(imgt2, (512, 512))
            feature = torchvision.transforms.functional.crop(imgt2, *rect)
            label = torchvision.transforms.functional.crop(img2, *rec)
            feature.save(pic_save + f'{i}' + '.png')
            label.save(label_save + f'{i}' + '.png')



if __name__ == "__main__":
    main()