模型训练技巧——Random Erasing

论文：https://arxiv.org/pdf/1708.04896v2.pdf

代码：https://github.com/zhunzhong07/Random-Erasing

1. 论文核心

 

Caption

          训练模型时，随机选取一个图片的矩形区域，将这个矩形区域的像素值用随机值或者平均像素值代替，产生局部遮挡的效果。该数据增强可以与随机切除、随机翻转等数据增强结合起来使用。

2. 代码实现

      代码中：

      Sl、Sh分别是需要随机擦除的矩形面积大小上下阈值；

      r1是限制矩形长宽比r的阈值 ；

import random
import math
import cv2
import numpy as np

class RandomErasing:
    """Random erasing the an rectangle region in Image.
    Class that performs Random Erasing in Random Erasing Data Augmentation by Zhong et al.

    Args:
        sl: min erasing area region 
        sh: max erasing area region
        r1: min aspect ratio range of earsing region
        p: probability of performing random erasing
    """

    def __init__(self, p=0.5, sl=0.02, sh=0.4, r1=0.3):

        self.p = p
        self.s = (sl, sh)
        self.r = (r1, 1/r1)
    

    def __call__(self, img):
        """
        perform random erasing
        Args:
            img: opencv numpy array in form of [w, h, c] range 
                 from [0, 255]
        
        Returns:
            erased img
        """

        assert len(img.shape) == 3, 'image should be a 3 dimension numpy array'

        if random.random() > self.p:
            return img
        
        else:
            while True:
                Se = random.uniform(*self.s) * img.shape[0] * img.shape[1]
                re = random.uniform(*self.r) 

                He = int(round(math.sqrt(Se * re)))
                We = int(round(math.sqrt(Se / re)))

                xe = random.randint(0, img.shape[1])
                ye = random.randint(0, img.shape[0])

                if xe + We <= img.shape[1] and ye + He <= img.shape[0]:
                    img[ye : ye + He, xe : xe + We, :] = np.random.randint(low=0, high=255, size=(He, We, img.shape[2]))

                    return img


if __name__ == "__main__":
    img = cv2.imread("test.jpg")
    RE = RandomErasing(p=1)
    for i in range(20):
        img1 = RE(img.copy())
        cv2.imshow("test", img1)
        cv2.waitKey(1000)  

3. 实验效果

Caption

        从实验可以看出，Random Erasing 可以降低识别的错误率，与Random flipping、Random cropping结合起来用效果更好。

Caption

         从实验可以看出，用上Random erasing可以提高目标检测的map。