NTIRE 2020 challenge on spectral reconstruction from an RGB image

Arad, Boaz and Timofte, Radu and Ben-Shahar, Ohad and Lin, Yi Tun and Finlayson, Graham and Givati, Shai and Li, Jiaojiao and Wu, Chaoxiong and Song, Rui and Li, Yunsong and Liu, Fei and Lang, Zhiqiang and Wei, Wei and Zhang, Lei and Nie, Jiangtao and Zhao, Yuzhi and Po, Lai Man and Yan, Qiong and Liu, Wei and Lin, Tingyu and Kim, Youngjung and Shin, Changyeop and Rho, Kyeongha and Kim, Sungho and Zhu, Zhiyu and Hou, Junhui and Sun, He and Ren, Jinchang and Fang, Zhenyu and Yan, Yijun and Peng, Hao and Chen, Xiaomei and Stiebel, Tarek and Koppers, Simon and Merhof, Dorit and Gupta, Honey and Mitra, Kaushik and Fubara, Biebele Joslyn and Sedky, Mohamed and Dyke, Dave and Banerjee, Atmadeep and Palrecha, Akash and Sabarinathan, Sabarinathan and Uma, K. and Vinothini, D. Synthiya and Sathya Bama, B. and Md Mansoor Roomi, S. M.; (2020) NTIRE 2020 challenge on spectral reconstruction from an RGB image. In: Proceedings - 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2020. IEEE Computer Society Press, USA, pp. 1806-1822. ISBN 9781728193601 (https://doi.org/10.1109/CVPRW50498.2020.00231)

[thumbnail of Arad-etal-IEEE-2020-NTIRE-2020-challenge-on-spectral-reconstruction-from-an-RGB]
Text. Filename: Arad_etal_IEEE_2020_NTIRE_2020_challenge_on_spectral_reconstruction_from_an_RGB.pdf
Accepted Author Manuscript

Download (5MB)| Preview


This paper reviews the second challenge on spectral reconstruction from RGB images, i.e., the recovery of whole- scene hyperspectral (HS) information from a 3-channel RGB image. As in the previous challenge, two tracks were provided: (i) a "Clean" track where HS images are estimated from noise-free RGBs, the RGB images are themselves calculated numerically using the ground-truth HS images and supplied spectral sensitivity functions (ii) a "Real World" track, simulating capture by an uncalibrated and unknown camera, where the HS images are recovered from noisy JPEG-compressed RGB images. A new, larger-than-ever, natural hyperspectral image data set is presented, containing a total of 510 HS images. The Clean and Real World tracks had 103 and 78 registered participants respectively, with 14 teams competing in the final testing phase. A description of the proposed methods, alongside their challenge scores and an extensive evaluation of top performing methods is also provided. They gauge the state-of-the-art in spectral reconstruction from an RGB image.