NTIRE 2020 Real World Super-Resolution Challenge - Track 1: Image Processing artifacts Forum

Go back to competition Back to thread list Post in this thread

> Dataset issues

Hi,

I was trying to download the data for the following https://competitions.codalab.org/competitions/22220#participate
I realized that links provided for Source Domain and Target Domain have different number of files, namely

Source Domain: Image Processing artifact
tr-x.zip is 14.0GB has 2650 files (Flickr2K)
https://data.vision.ee.ethz.ch/alugmayr/NTIRE2019/public/Corrupted-tr-x.zip

Target Domain: Clean High Quality Image
tr-y.zip is 3.3GB and has 800 files (DIV2k)
https://data.vision.ee.ethz.ch/alugmayr/NTIRE2019/public/Corrupted-tr-y.zip

Should it not be that both have the same number of images and from the same dataset? Please let me know if I am downloading incorrect links.

Thanks
Sunando

Posted by: sunandosengupta @ Jan. 6, 2021, 2:57 p.m.

Hi,

Please check carefully the description of the challenge.
Also you can check the challenge report:
http://openaccess.thecvf.com/content_CVPRW_2020/papers/w31/Lugmayr_NTIRE_2020_Challenge_on_Real-World_Image_Super-Resolution_Methods_and_Results_CVPRW_2020_paper.pdf

This is a challenge asking you to learn from unpaired data. Is it not the fully supervised learning with input-output image pairs commonly found in the SR literature.

Best,
Radu

Posted by: Radu @ Jan. 6, 2021, 3:04 p.m.

Thanks Radu.

Posted by: sunandosengupta @ Jan. 6, 2021, 3:32 p.m.
Post in this thread