Submission instructions and utility files now available here:
https://github.com/boazarad/NTIRE2022_spectral
Please note that there has been an issue with the training data, correct, 16-channel training files have been provided on the project homepage.
Please update your training files.
Posted by: boazarad @ Feb. 7, 2022, 1:06 a.m.Thanks for the update.
The MSFA->Mosaic conversion is still confusing though. Based on the code "mosaic *= (TYPICAL_SCENE_REFLECTIVITY / mosaic.mean()) * MAX_VAL_12_BIT", MSFA files of different brightnesses will be converted to the same Mosaic file. In this case, shouldn't SAM be a better metric for assessment instead of PSNR?
Posted by: divx @ Feb. 8, 2022, 12:49 a.m.Both PSNR and SAM are reported, and a case can be made for the advantages and disadvantages of each of them. Based on our literary review, we’ve selected PSNR as our primary metric, but SAM will be measured and reported for all submission in the final report - if there is a significant difference in PSNR/SAM rankings across submissions, this will definitely be a point for discussion.
Posted by: boazarad @ Feb. 8, 2022, 11:29 p.m.Manual leaderboard now available here:
https://docs.google.com/spreadsheets/d/1MXwyVopb-lwQox-qfeLORB01Z3fIzb1w0oLWMc0LFQQ/edit?usp=sharing
Will be updated every 24-48 hours
Posted by: boazarad @ March 19, 2022, 3:11 p.m.