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

Low-Light Experimental settings #10

Closed
Allen5123 opened this issue Mar 10, 2024 · 9 comments
Closed

Low-Light Experimental settings #10

Allen5123 opened this issue Mar 10, 2024 · 9 comments

Comments

@Allen5123
Copy link

您好,請問Low-Light Experimental settings 中的 Loss = L_{res}與Sampleing method = SM-Res該如何設置?
目前嘗試以下配置會error
objective = "pred_res"
test_res_or_noise = "res"

image
@nachifur
Copy link
Owner

您可以先自行修改和调试一下代码。我们正在逐步地整理和更新代码。

@nachifur
Copy link
Owner

注意:对于低光照,需修改train.pyequalizeHist=True

@Allen5123
Copy link
Author

謝謝,這部分有注意到,會持續關注後續更新

@Allen5123
Copy link
Author

您好,想請問一些deraining任務的配置 :

  1. sampling均為res + noise 如下 ?
    objective = 'pred_res_noise'
    test_res_or_noise = "res_noise"
  2. convert_to_ddim 需要改成False,才會套用Eq(14)的scheduling ?
  3. 原文中P(1-x,1)對應的是linear decreasing還是P(x,1) ?
    image
    4.目前source code中的alphas / beta的schedule是分別對應到Image restoration中的P(1-x,1)與P(x,1)嗎,還是需要另作更改 ?
    image
  4. Training set是只使用Raindrop/train中的861張image ?

@nachifur
Copy link
Owner

@Allen5123

  1. Yes, for image restoration, convert_to_ddim=False.

  2. For sampling, L_{res} + L_{\epsilon} (one Adam) needs to modify the loss function. The current code is L_{res}, L_{\epsilon} (two RAdma).

  3. For P(1-x,1), see: pre-trained ckpt #12 (comment)

  4. only Raindrop/train for training.

@Allen5123
Copy link
Author

您好,謝謝回復!

@Allen5123
Copy link
Author

您好,想請問關於image restoration任務的問題:

  1. 如果已將line 640改為convert_to_ddim=False,是否也須一併更改line 726至convert_to_ddim=False ?
  2. 承上,Testing時,line 771 ~ 781這邊的設置與訓練時,使用的alpha, beta scheduling(__init__內初始化的值)不一樣,理論上,它們應該套用同樣的alpha, beta scheduling ? (e.g.相同的sum_scale, ratio, schedule)

@Allen5123 Allen5123 reopened this May 9, 2024
@nachifur
Copy link
Owner

您好,想請問關於image restoration任務的問題:

  1. 如果已將line 640改為convert_to_ddim=False,是否也須一併更改line 726至convert_to_ddim=False ?
  2. 承上,Testing時,line 771 ~ 781這邊的設置與訓練時,使用的alpha, beta scheduling(__init__內初始化的值)不一樣,理論上,它們應該套用同樣的alpha, beta scheduling ? (e.g.相同的sum_scale, ratio, schedule)

是的,convert_to_ddim = False in L726

@Allen5123
Copy link
Author

Thanks~

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