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

Why is loss rising #11

Open
zqnnn opened this issue Mar 26, 2020 · 10 comments
Open

Why is loss rising #11

zqnnn opened this issue Mar 26, 2020 · 10 comments

Comments

@zqnnn
Copy link

zqnnn commented Mar 26, 2020

I trained on my own dataset, the reg loss is rising , and loss is a negative number. Why does this happen?

@zqnnn
Copy link
Author

zqnnn commented Mar 27, 2020

请问一下,我想在自己的数据上训练,是不是应该先训练基础的Sphere64,再固定之前的训练不确定模块?Sphere64网络的训练有代码可以参考吗?

@Ontheway361
Copy link

从数学公式上来看,negMLS不能保证非负。老铁,加个微信讨论下细节哈,我wechat:On_theway24,备注一下是PFE模型讨论哈。

@twmht
Copy link

twmht commented Apr 29, 2020

@Ontheway361

從 PAPER 的 Equation(3) 來看,會導致出現負號的原因是因為加了 log 嗎

@twmht
Copy link

twmht commented Apr 29, 2020

只要可以保證 log(σ2(l)i+σ2(l)j) 為正的話,基本上就是 negMLS 就是恆正了

@Ontheway361
Copy link

@twmht , 事实上,不能保证 log(σ2(l)i+σ2(l)j) > 0!

@twmht
Copy link

twmht commented May 6, 2020

@Ontheway361

我覺得應該 log 裡面要加個 exp?

@Ontheway361
Copy link

坦白来说,我认为论文中的公式(3)是错误的,我自己没有推导出来,然后发邮件咨询作者,作者也不回复邮件。所以,我觉得这篇文章就是垃圾,推荐你看CVPR-2020的一篇paper :Data Uncertainty Learning in Face Recognition.

@twmht
Copy link

twmht commented May 11, 2020

@Ontheway361

那篇我知道呀! 但目前沒有 source code 不好複製

@twmht
Copy link

twmht commented May 11, 2020

PFE 的 similarity 沒有上限,但作者似乎也沒給一個實驗建議值

@hungsing92
Copy link

推不出来公式3+1

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

4 participants