positive666 / yolo_research

based on yolo-high-level project (detect\pose\classify\segment\):include yolov5\yolov7\yolov8\ core ,improvement research ,SwintransformV2 and Attention Series. training skills, business customization, engineering deployment C
GNU General Public License v3.0
758 stars 146 forks source link

Improvement of the training section #7

Closed petter-Wang closed 2 years ago

petter-Wang commented 2 years ago

❔Question

您好,之前跟您在博客上沟通过,看到您开了ISSUE,我按照您的改进思路,想在LOSS和标签匹配上做一下改进,但是我现在没有什么思路,您能再分享下或者一些改进这部分代码上的工作吗?

Additional context

positive666 commented 2 years ago

❔Question

您好,之前跟您在博客上沟通过,看到您开了ISSUE,我按照您的改进思路,想在LOSS和标签匹配上做一下改进,但是我现在没有什么思路,您能再分享下或者一些改进这部分代码上的工作吗?

Additional context

这部分工作我也没做 你可以参考下SIMOTA 目前的方法还是以IOU的基准来定义,LOSS的话我做个实验在更新把 网上有很多 Focal的变体 之前那个EIOULOSS 属实是反向优化了

petter-Wang commented 2 years ago

❔Question

您好,之前跟您在博客上沟通过,看到您开了ISSUE,我按照您的改进思路,想在LOSS和标签匹配上做一下改进,但是我现在没有什么思路,您能再分享下或者一些改进这部分代码上的工作吗?

Additional context

这部分工作我也没做 你可以参考下SIMOTA 目前的方法还是以IOU的基准来定义,LOSS的话我做个实验在更新把 网上有很多 Focal的变体 之前那个EIOULOSS 属实是反向优化了

感谢回答 后续还请多多指教 大佬C++ 强啊学习了

github-actions[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.