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

question about "window size" and "striped size" #14

Open
wangjg33 opened this issue Apr 17, 2023 · 1 comment
Open

question about "window size" and "striped size" #14

wangjg33 opened this issue Apr 17, 2023 · 1 comment

Comments

@wangjg33
Copy link

Congratulations, this job is very outstanding! But when I read your code and paper, there are two questions.

  1. Where are the parameter settings for the final three models of different sizes placed? Is it the parameter settings in the ./config/model/grl/grl_base.yaml or ./models/networks/grl.py ?
    The parameter settings in these two files seem different, especially the settings for "stripe_shift".

  2. Second question is about the two parameter settings for "stripe_size" and "stripe_groups". In your paper, you mentioned the anisotropic images features, which I think is a good idea. But when reading the code, I found that in the grl.py file, the default “stripe_size” is set to [8, 8], and the “stripe_groups” is set to [None, None], which seems to be equivalent to a window attention with a window size equal to 8.
    However, in the base.yaml file, they are set [8, null] and [null, 4] , respectively. Through this setting, it can generate a rectangular window, but it seems that there is only one direction of anisotropy, which does not reflect the vertical and horizontal anisotropy shown in Figure 4 of your paper.

@ofsoundof
Copy link
Owner

Thanks a lot for your questions.

The two paramters are actually overriden under .config/experiment.

Each task might have different settings. For example, for image denoising, the window size is set here. And the stripe size is set here.

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