-
Notifications
You must be signed in to change notification settings - Fork 2
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
你好,请教关于人脸检测的问题 #9
Comments
主要是充分的增益和训练. 匹配策略有差异但是应该不是重点! |
我归纳一下不同之处是旋转增益+匹配策略。你说匹配策略不是重点,那么重点就是旋转增益了? |
1 similar comment
我归纳一下不同之处是旋转增益+匹配策略。你说匹配策略不是重点,那么重点就是旋转增益了? |
增益的尺度分布非常影响 val 集的结果, 如果增益的人脸尺度和val集接近, 验证集的得分会更高! 这是一种刷分策略. 另外对于小模型, 应当使得人脸尺度分布在其能力范围之内, 使之能够更有效地被训练. 毕竟小模型的能力是有限的! 旋转增益应当说是工程上实现的选择, 对于精度影响我没有看到显著的影响. 该做法是为了我们落地时能够支持更自由的人脸角度的检测. widerface 的验证集, 人脸朝向大部分为正脸, 不符合我们部署的现实场景. bonus: 匹配这个点, 如果能够使用带角度的人脸框匹配rotated_boxes, 精度还能进一步提升. |
我发现你基于Face-Detector-1MB-with-landmark 的改进版本准确率有所提高,
想了解你做了哪些地方的优化呢?谢谢!
The text was updated successfully, but these errors were encountered: