PSNR evaluation used in CodaLab is "skimage.metrics" library
https://newcodalab.lri.fr/prod-private/dataset_data_file/None/394f3/main.py?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Expires=86400&X-Amz-SignedHeaders=host&X-Amz-Signature=9f0dc58884dd08258944b80cde3555be93720999377a2974279891f088285f7d&X-Amz-Date=20200316T073144Z&X-Amz-Credential=AZIAIOSAODNN7EX123LE%2F20200316%2Fnewcodalab%2Fs3%2Faws4_request
However in github repo, you calculate the formula directly and discard border pixels.
https://github.com/thstkdgus35/ntire-2020-deblur-mobile/blob/master/metric.py
Which one will be used in final scoring?
Thank you.
Posted by: jimmy.chiang @ March 16, 2020, 7:38 a.m.Dear jimmy.chiang,
The evaluation will use functions from skimage.metrics library including the boundary pixels.
The GitHub code will not be used for the final evaluation.
I'm sorry for the inconsistency and we will fix it.
Best,
Seungjun