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

关于论文里的loss一些问题 #18

Closed
mxs30443 opened this issue Jul 10, 2019 · 3 comments
Closed

关于论文里的loss一些问题 #18

mxs30443 opened this issue Jul 10, 2019 · 3 comments

Comments

@mxs30443
Copy link

首先感谢作者的分析,无论是论文还是源码都是很好的工作。向你学习!

image
但是我阅读过您的论文后心里存在一些疑问,在您的loss函数上的设计我有些疑问 :

您loss的设计主要是想用辅助网络的角度去辅助学习landmark的位置,但是根据公式如果辅助网络学习得到的角度和groundtruth相差小的话(极端点,差值直接为0)那么上述loss应该为0,

假如角度学习的3个值比landmark要学习136个值更加容易学习更容易过拟合,那么会不会存在角度辅助网络反而阻挠了landmark的学习。而且您没有给出不使用该loss的对比试验结果,所以我有点疑问。

心中只是有这个疑问,不知道是不是我论文没看明白没彻底理解作者意思。希望作者在闲暇之余帮忙解答

@guoqiangqi
Copy link
Owner

我不是作者。。不过你说的欧拉角预测损失为0基本不会出现,分支网络的训练会加强网络公共部分的信息提取能力,不同分支再针对不同任务学习特征。

@guoqiangqi
Copy link
Owner

感兴趣的话你可以试试没有分支网络的哈

@mxs30443
Copy link
Author

恩恩,我去试试,谢谢你的解答!

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

No branches or pull requests

2 participants