当前位置:网站首页>mAPH - Waymo dataset
mAPH - Waymo dataset
2022-08-11 06:16:00 【zhSunw】
mAPH
I believe you have read itWaymoStudents who test papers on the dataset will find itmAPH这个指标,But I checked the information on the whole network and there is no relevant explanation(Except you need to climb the wall to enterWaymo官网下的What’s Next).
So in order to help the next person who wants to understand but can't find the relevant explanation,Here is the official introduction screenshot and my understanding(其实不难)
First, a screenshot of the official website is given,It is good if the students with better foundation can read it by themselves:
首先我们回顾一下 m A P mAP mAP:计算出 P r e c i s i o n Precision Precision与 R e c a l l Recall Recallafter two parameters,以 P r e c i s i o n Precision Precision和 R e c a l l Recall Recallare the vertical and horizontal axes, respectively,就可以画出 P r e c i s i o n − R e c a l l ( P − R ) Precision-Recall(P-R) Precision−Recall(P−R)曲线,如图:
P − R P-R P−RThe area enclosed by the curve is called A v e r a g e P r e c i s i o n ( A P ) Average Precision(AP) AveragePrecision(AP).
其中:
P r e c i s i o n = T P / ( T P + F P ) Precision=TP/(TP+FP) Precision=TP/(TP+FP)
R e c a l l = T P / ( T P + F N ) Recall=TP/(TP+FN) Recall=TP/(TP+FN)
至于 T P 、 T N 、 F P 、 F N TP、TN、FP、FN TP、TN、FP、FNThe sum of the quantities of the four parameters is the total sample size,The classification results they represent are shown in the table below:
Finally, we look back to the official website to explain,不难知道,它的计算方法与mAPThe difference lies in the drawingP-RThe graph is predicted by headingAccuracy weighted,即TPparameters will be predictedheading加权.predictedheading与真实的headingThe angle difference is A(A值域为 [ 0 , Π ] [0,Π] [0,Π]),for every real exampleTP计算方法如下:
T P = 1 − A / Π TP = 1 - A/Π TP=1−A/Π
( m A P mAP mAPof each target T P TP TP为1,即混淆矩阵中TPThe number of and here becomes the weighted for each targetTP之和)
All other indicators remain unchanged,Draw this way P − R P-R P−Rfigure and calculatemAP就是 m A P H mAPH mAPH了.
边栏推荐
猜你喜欢
随机推荐
梅科尔工作室-DjangoWeb 应用框架+MySQL数据库第六次培训
BGP联邦实验
GBase 8s分片技术介绍
梅科尔工作室-PR第三次培训笔记(效果与转场及插件使用)
GBase 8s的分片和索引
OSPF综合实验
《现代密码学》学习笔记——第三章 分组密码 [二] AES
内核与用户空间通过字符设备通信
华为adb wifi调试断线问题解决
恶劣天气 3D 目标检测数据集收集
GBase 8a MPP Cluster产品高级特性
安全帽识别系统-解决监管难题
lvs的keepalived
若依分离版—增加通知公告预览功能
Pay “Attention” to Adverse Weather
LNMP源码搭建
动画(其二)
架构设计杂谈
TAMNet:A loss-balanced multi-task model for simultaneous detection and segmentation
Waymo数据集使用介绍(waymo-open-dataset)