search

RaymondLZhou / deepDeband

Official repository of the Deep Image Debanding conference paper. This research project was done with Dr. Shahrukh Athar, Zhongling Wang, and Prof. Zhou Wang, and the work is published in ICIP 2022.
MIT License
15 stars 4 forks source link

Some questions about details of dataset #1

Closed csxyhe closed 9 months ago

csxyhe commented 1 year ago

I have downloaded your dataset. It's an excellent dataset. But I have some questions on it.

  1. The dataset only includes 51420 pairs, but not 51490. Are there some mistakes?
  2. Could you give some specific tips on how to partition the whole dataset into training, validation, testing three parts as your paper's description? I mean to partition it as Table 1 given in your paper. I would be honored if you could reply to me. Thanks!
RaymondLZhou commented 1 year ago

Thanks for reaching out!

  1. The dataset there should be the most up to date I have - not sure why there's a difference, but what's uploaded is the most recent
  2. I'd recommend partitioning it so that for any image, all of its patches are either in the training, validation, or test set (we don't want patches from the same image in both training and validation, for example). Other than that, the images are pretty diverse in terms of visual content so there should be a good balance already regardless of how you split the patches

On Mon, Mar 20, 2023 at 9:01 AM yogart-he @.***> wrote:

I have downloaded your dataset. It's an excellent dataset. But I have some questions on it.

  1. The dataset only includes 51420 pairs, but not 51490. Are there some mistakes?
  2. Could you give some specific tips on how to partition the whole dataset into training, validation, testing three parts as your paper's description? I mean to partition it as Table 1 given in your paper. I would be honored if you could reply to me. Thanks!

— Reply to this email directly, view it on GitHub https://github.com/RaymondLZhou/deepDeband/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOGJJMT2KM63J4RJ2EJWDNTW5BILPANCNFSM6AAAAAAWBBXIP4 . You are receiving this because you are subscribed to this thread.Message ID: @.***>