-
Notifications
You must be signed in to change notification settings - Fork 52
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
Reproduce LD in yolov8? #59
Comments
大佬,最近又反复调了下,之前选择学生模型并没有学习能力,现在换了大一点模型,目前是涨一个点,发现一个奇怪的问题,学生模型太关注老师模型的精准度,召回率我这边考虑将剩下的负样本带进去蒸馏看看,不知道还有哪些可借鉴 |
|
好的 我准备在VOC试了有效再搞COCO,tal是正负样本匹配的方式主要没有阈值概念,可能时半会还没有想出获取valuable negtive region办法,我们只蒸馏正样本区域,并没有考虑到怎么召回更多的围绕再gt周围的anchor |
hhh 这么刺激,我试一下 |
大佬,我最近复现特征蒸馏非常不顺利(yolov8上),能加个微信聊聊吗,我发过您邮箱 |
没收到你的邮件, 2668129173是你的qq吗 |
大佬有空的适合加一下,我的微信号wangtao2668129173 |
大佬我最近也在v8上搞蒸馏,能否参考下是如何添加蒸馏部分的 代码 |
@wangtao2668129173 大佬,您现在yolov8蒸馏搞得咋样,想跟您学习下 |
|
大佬我加你微信了 |
感谢您的工作,在这里我有几个问题想和您交流以下:
①我目前在yolov8上采用VOC数据集分别训练学生和老师模型,分别选择pos位置的cls和reg distribution分别为main kd,main ld,如果按照原论文设置会导致kd loss特别大,因此我设置kd参数为weight: 0.25,T=2,ld依然是0.25,T10,模型只涨了0.6%
②yolov8由于适用TAL,对于valuable region loss不太好选择
问题1我尝试调了几次参数就是不太work
请问大佬有啥意见吗
The text was updated successfully, but these errors were encountered: