CC-NORD: A Camera-Invariant Global Color Constancy Dataset

Diclehan Ulucan, Oguzhan Ulucan, Marc Ebner

Abstract: Color constancy benchmarks are mostly created by using cameras with similar hardware and by capturing the scenes under similar illuminants. Therefore, ground truth information is affected by the cameras' sensor response characteristics, and during the evaluation of color constancy algorithms, some illuminants are neglected. In this study, we introduce a large-scale synthetic color constancy dataset called CC-NORD. Since CC-NORD is formed by using computer graphics, the ground truth information is independent of the capturing device. Also, to provide images created under a wide range of illuminants, the scenes in CC-NORD are rendered with a high number of distinct illuminants including lights having values away from the color temperature curve. To the best of our knowledge, this is the first color constancy dataset containing scenes rendered under illuminants on the edge and outside the color temperature curve. Upon publication, CC-NORD will be publicly available on the first author's webpage.