search

c-yn / IRNeXt

[ICML2023] IRNeXt: Rethinking Convolutional Network Design for Image Restoration
MIT License
48 stars 4 forks source link

datasets #3

Closed Urian-oy closed 6 months ago

Urian-oy commented 6 months ago

As we kown, in the OTS dataset, a clear image corresponds to many foggy images. How to process images on the dataset folder to make the program run

c-yn commented 6 months ago

Hi, please refer to Line 64 of Dehazing/OTS/data/data_load.py.

Urian-oy commented 6 months ago

Hi, please refer to Line 64 of Dehazing/OTS/data/data_load.py.

Thanks!I have already done.There is a para(ps=256) need to seu up . Dehazing/OTS/data/data_load.py--->line 10: DeblurDataset(image_dir, ps=128),

Urian-oy commented 6 months ago

Hi, please refer to Line 64 of Dehazing/OTS/data/data_load.py.

By the way , why I set 'ps' other number, it doesnot work.There are some magic?

c-yn commented 6 months ago

The ps number must be divided by 4.

Urian-oy commented 6 months ago

The ps number must be divided by 4.

Thanks a lot.I will try it.Have a nice day

c-yn commented 6 months ago

Thanks. Also may be 32. It depends on the downsampling rate of the model.