当前位置:网站首页>yolov5训练可视化指标的含义
yolov5训练可视化指标的含义
2022-07-01 08:30:00 【余生的观澜】
背景
在v5的训练过程中,使用wandb是一个很常见可视化工具,它有很多的图表,可以对训练过程中可以提取的评价指标进行可视化,今天做一个整理。
可视化指标
LOSS
loss分为cls_loss, box_loss, obj_loss三部分。
cls_loss用于监督类别分类,计算锚框与对应的标定分类是否正确。
box_loss用于监督检测框的回归,预测框与标定框之间的误差(CIoU)。
obj_loss用于监督grid中是否存在物体,计算网络的置信度。
参考资料:https://blog.csdn.net/u011994454/article/details/119564834
metrics
mAP([email protected]),这是一个对检测能力要求更高的标准。
mAP([email protected]),跟Pascal VOC mAP标准计算方式一致;
mAP([email protected][0.5:0.05:0.95]),需要计算10个IoU阈值下的mAP,然后计算平均值。这个评估指标比仅考虑通用IoU阈值(0.5)评估指标更能体现出模型的精度。
https://www.jianshu.com/p/fd9b1e89f983
边栏推荐
- 《单片机原理及应用》—定时器、串行通信和中断系统
- vscode自定义各个区域的颜色
- leetcode T31:下一排列
- [untitled]
- Airsim radar camera fusion to generate color point cloud
- seaborn clustermap矩阵添加颜色块
- Leetcode t39: combined sum
- When using charts to display data, the time field in the database is repeated. How to display the value at this time?
- Yolov5 advanced 7 target tracking latest environment setup
- CPU design practice - Chapter 4 practical tasks - simple CPU reference design and debugging
猜你喜欢

The era of low threshold programmers is gone forever behind the sharp increase in the number of school recruitment for Internet companies

The data analyst will be ruined without project experience. These 8 project resources will not be taken away

7-26 word length (input and output in the loop)

CPU design practice - Chapter 4 practical tasks - simple CPU reference design and debugging

Field agricultural irrigation system

《微机原理》—总线及其形成

Properties of 15MnNiNbDR low temperature vessel steel, Wugang 15MnNiDR and 15MnNiNbDR steel plates

【C】 Summary of wrong questions in winter vacation

Huawei machine test questions column subscription Guide

分享2022上半年我读过的7本书
随机推荐
機動目標跟踪——當前統計模型(CS模型)擴展卡爾曼濾波/無迹卡爾曼濾波 matlab實現
Provincial election + noi Part VI skills and ideas
Audio audiorecord create (I)
Redis publish subscription
【C】 Summary of wrong questions in winter vacation
2022 ordinary scaffolder (special type of construction work) examination question bank and the latest analysis of ordinary scaffolder (special type of construction work)
Tita OKR: a dashboard to master the big picture
为什么LTD独立站就是Web3.0网站!
Matlab [functions and images]
Leetcode t29: divide two numbers
MAVROS发送自定义话题消息给PX4
《单片机原理及应用》—定时器、串行通信和中断系统
To prevent "activation" photos from being muddled through, databao "live detection + face recognition" makes face brushing safer
MATLAB小技巧(16)矩阵特征向量特征值求解一致性验证--层次分析
Use threejs simple Web3D effect
Analysis of slice capacity expansion mechanism
Vscode customize the color of each area
Provincial election + noi Part VII computational geometry
How can beginners correctly understand Google's official suggested architectural principles (questions?)
Using settoolkit to forge sites to steal user information