NTIRE 2022 Spectral Demosaicing Challenge and Data Set

Abstract

This paper presents the first challenge on demosaicing of natural spectral images for snapshot hyperspectral imaging systems (HIS) which utilize a multi-spectral filer array (MSFA), i.e., the recovery of whole-scene hyperspectral information from spatially sub-sampled hyperspectral information. This challenge expands the "ARAD_1K" data set to a first-of-its-kind large-scale data set for multispectral filter array demosaicing of natural scenes containing 1,000 images. Challenge participants were required to recover hyperspectral information from synthetically generated MSFA images simulating capture by a known calibrated snapshot mosaic hyperspectral camera. The challenge was attended by 157 teams, with 29 teams competing in the final testing phase, 7 of which provided detailed descriptions of their methodology which are included in this report. The performance of these submissions is reviewed and provided here as a gauge for the current state-of-the-art in multi-spectral filter array demosaicing of natural images.