pytorch c++opencv数据预处理

目录

裁剪，遮挡，数据增强： 

人脸识别增强:

opencv需要转pil

减均值，除方差示例。

3种预处理比较

c++ opencv mat 减均值，除方差

c++ 减均值除方差方法2

opencv转pil，再转torch，数据没有变化，pil不会自动变为rgb格式，

---

裁剪，遮挡，数据增强： 

import torch as t
import numpy as np
import torchvision as tv
import matplotlib.pyplot as plt
from torchvision import transforms
from torchtoolbox.transform import Cutout
 
ROOT = '../pytorch/cifar-10'
BATCH_SIZE=128

train_transform = transforms.Compose([
    transforms.RandomCrop(32, padding=4),#每边填充4，把32^*32填充至40*40，再随机裁剪
    Cutout(0.5),#参数是遮挡的概率
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor()
])

人脸识别增强:

  self.transform = transform