Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

训练YOLOv8 #103

Closed
1 of 2 tasks
WanShaw opened this issue Mar 9, 2023 · 4 comments
Closed
1 of 2 tasks

训练YOLOv8 #103

WanShaw opened this issue Mar 9, 2023 · 4 comments
Labels
question Further information is requested

Comments

@WanShaw
Copy link

WanShaw commented Mar 9, 2023

问题确认 Search before asking

  • 我已经查询历史issue,没有类似需求。I have searched the issues and found no similar feature requests.

需求描述 Feature Description

想请问一下,什么时候可以实现训练yolov8呢?我觉得pdyolo的实现比官方或者mmyolo更好,所以一直想等着用用!

是否愿意提交PR Are you willing to submit a PR?

  • Yes I'd like to help by submitting a PR!
@nemonameless
Copy link
Collaborator

谢谢关注。由于工作安排问题,我们是将YOLOv8的训练实现放到了 飞桨黑客松论文复现赛 里,具体比赛可看PaddlePaddle Hackathon 飞桨黑客马拉松第四期。我们也会及时评估选手的进度,必要时辅助选手或自行实现,由于比赛原因预计4月底才可以公布实现。

就当前性能来看,PPYOLOE+ 模型只训80epoch的情况下还是可以略胜YOLOv8的500epoch,你可以训更多epoch试试。

此外PPYOLOE+的小目标、旋转框、密集检测、半监督检测、检测蒸馏、tiny版本、obj365预训练、transformer等超大型backbone、各种人车工业垂类应用都很齐全,欢迎使用。
https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyoloe
https://github.com/PaddlePaddle/PaddleDetection/releases/tag/v2.6.0

@WanShaw
Copy link
Author

WanShaw commented Mar 9, 2023

好的,谢谢回复。请问有ppyoloe+直接在coco数据集上的训练日志吗?
还有ppyoloe的+版本,是在论文预印后改动的,我好像没有在issue中发现’在backbone中block分支中增加alpha参数‘的解释,该参数的具体目的,请相关大佬能解释一下吗?这个参数对另一个并行分支进行了加权,而且alpha是可学习的。

@nemonameless
Copy link
Collaborator

ppyoloe+只要训80epoch相对已经很快了。后面会统一放出来log,都是每10epoch一次eval精度的。
alpha参数就是可学习参数因子,几乎不增加参数量计算量,也不影响速度,但是可以提高mAP,coco上虽不明显,但下游任务业务数据集上提升很明显。

@nemonameless nemonameless added the question Further information is requested label Mar 14, 2023
@WanShaw
Copy link
Author

WanShaw commented Mar 14, 2023

好的,感谢大佬回答

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

2 participants