search

Algolzw / image-restoration-sde

Image Restoration with Mean-Reverting Stochastic Differential Equations, ICML 2023. Winning solution of the NTIRE 2023 Image Shadow Removal Challenge.
https://algolzw.github.io/ir-sde/index.html
MIT License
576 stars 42 forks source link

Dataset preparation #19

Open ensyvz opened 1 year ago

ensyvz commented 1 year ago

How should a custom dataset structured? For example, i have a gt image and an image with blur applied at random places. Should i also have the mask for blur regions? And how should i structure this dataset? Thanks for your efforts.

Algolzw commented 1 year ago

Hi! If your custom dataset consists of paired images, you can just make two directories (LQ and GT). Currently, our code doesn't require prior information (such as masks) and thus you can structure the dataset by only considering LQ and HQ images.

But if you really want to use the mask information, you can also put all masks in a separate directory like the bokeh transform config and dataloader.

Best.