计算两张图片的IOU

没有出口的猎户座

已于 2022-04-17 02:24:08 修改

阅读量2.1k

点赞数 3

分类专栏： My Project 文章标签： pytorch 深度学习 python

于 2022-02-12 05:24:58 首次发布

本文链接：https://blog.youkuaiyun.com/anelance/article/details/122891755

版权

My Project 专栏收录该内容

19 篇文章

订阅专栏

这篇博客介绍了一段Python代码，用于计算两张图像的交并比（Intersection over Union, IoU）得分。首先读取两张图像，然后将它们从uint8类型转换为整数，并进行减法操作得到差值数组。接着，通过布尔运算求得两图的交集和并集区域，从而计算IoU分数。代码适用于处理2D或3D图像，但需要注意图像的通道数。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

import cv2
import os
import numpy as np
from pathlib import Path
import time
from PIL import Image
import numpy as np

#Read images
img1 = np.array(Image.open(032439.413.jpg"))#标注图
img2 = np.array(Image.open(32439.413.png"))#pytorch图片
#print(img1.shape)
#print(img2.shape)


#convert unit8 to int numbers, then subtract, then 2D to 1D array
#subArr = np.subtract(img1[:, :, 0].tolist(), img2.tolist()).flatten()

intersection = np.sum(np.logical_and(img1, img2[:, :, 0]))
union = np.sum(np.logical_or(img1, img2[:, :, 0]))


#Intersection over Union
iou_score = intersection / union

print(iou_score)

报错：

Img1 should be the .jpg file that has 3 channels

If your image is 2D, just delete one of the ,: from the code

Or just delete the [:,:,0]

If it's 2D, just img1, if it's 3D img1[:, :, 0]