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

请教关于生成结果图像质量下降等问题 #44

Closed
irine1210 opened this issue Nov 21, 2022 · 3 comments
Closed

请教关于生成结果图像质量下降等问题 #44

irine1210 opened this issue Nov 21, 2022 · 3 comments

Comments

@irine1210
Copy link

作者您好,作为一名初学者,您的工作让我受益良多。有几个问题想请教一下:

  1. 在复现论文的过程中,我发现在对Young个glasses这两个属性分别进行转换,训练到11万次左右的时候出现了如下情况(在对其他组合训练的时候则不会产生):请问这是由于什么原因导致的?

image

image

  1. 基于问题1,不同的标签组合是否会对训练结果产生影响?
  2. 是否训练的次数越多效果就一定越好呢?还是说最好的结果出现在哪里要通过生成的结果直观查看
  3. 在做reference-guided时是否可以通过标签对参考图像进行约束?

期待您的回复,感谢感谢!祝您科研顺利

@irine1210 irine1210 changed the title 关于生成结果图像质量下降等若干问题 请教关于生成结果图像质量下降等问题 Nov 22, 2022
@imlixinyang
Copy link
Owner

您好~回复迟了不好意思。

  1. 我猜测更多是因为对抗训练本身在后期崩了导致的,这里可能你得改变一下损失或者调大R1正则化权重。
  2. 一定会有影响,因为同时训练,网络会权衡任务的难易从而自动调整在每种任务上所用的自身资源。
  3. 并不一定,因为我也发现模型在后期reference-guided任务会越来越差(也是因为HiSD原本并没有引入reference-guided的路线),但足够稳定的训练一般最后的模型不会比训练过程中最好的模型差,特别是适用了EMA做模型平均后。
  4. 这个没太理解。

@irine1210
Copy link
Author

好的好的,感谢您的详细回复!让我看待HiSD有了全新的视角,我再继续研究一下

@imlixinyang
Copy link
Owner

不客气~ 有任何问题都可再交流

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