NTIRE 2025 Ambient Lighting Normalization Challenge Report

Florin-Alexandru Vasluianu, Tim Seizinger, Zhuyun Zhou, Zongwei Wu, Radu Timofte, Yuanfei Bao, Xingbo Wang, Xin Lu, Jiarong Yang, Anya Hu, Kunyu Wang, Jie Xiao, Dong Li, Xueyang Fu, Zheng-Jun Zha, Zihao Fan, Xi Wang, Yurui Zhu, Kean Liu, Senyan Xu, Hongjian Liu, Yupeng Xiao, David Serrano-Lozano, Francisco A. Molina-Bakhos, Danna Xue, Yixiong Yang, Maria Pilligua, Ramon Baldrich, Maria Vanrell, Javier Vazquez-Corral, Xuan Sun, Zijie Lou, Ting Liu, Kuldeep Purohit, Jameer Babu Pinjari, Yilin Zhang, Huan Zheng, Yanyan Wei, Suiyi Zhao, Shengeng Tang, Zhao Zhang, Yushen Zuo, Zongqi He, Zhe Xiao, Cuixin Yang, Rongkang Dong, Jun Xiao, Kin-Man Lam, Nikhil Akalwadi, Vijayalaxmi Ashok Aralikatti, Dheeraj Damodhar Hegde, Ramesh Ashok Tabib, Uma Mudenagudi, Anas M. Ali, Bilel Benjdira, Wadii Boulila; Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops, 2025, pp. 1289-1300

Abstract

This report presents an overview of the NTIRE 2025 Ambient Lighting Normalization Challenge, a competition designed to advance techniques for improving image consistency under varying lighting conditions. Participants were tasked with developing algorithms capable of normalizing images acquired under various direct lighting systems to ambient lighting equivalents while preserving image quality, detail, and color accuracy. With a total number of 171 participants, the first edition of the challenge resulted in a number of 10 Final Phase submissions, which are part of the challenge benchmark. Conditions such as image restoration fidelity and the perceptual quality of the normalized outputs form the base of the proposed ranking. A user study including subjects with various backgrounds, including professional photographers, is backing the proposed ranking, emphasizing clearly the top-performing solutions. This report outlines the competition framework, dataset composition, evaluation metrics, and performance of different approaches. The top-performing methods leveraged deep learning strategies, using both end-to-end learning techniques and solutions based on iterative refinement. A comparative analysis of submissions highlights the strengths and limitations of each approach, offering insights into the effectiveness of all proposed ambient lighting normalization strategies.

Related Material

[pdf] [supp]
[bibtex]
@InProceedings{Vasluianu_2025_CVPR, author = {Vasluianu, Florin-Alexandru and Seizinger, Tim and Zhou, Zhuyun and Wu, Zongwei and Timofte, Radu and Bao, Yuanfei and Wang, Xingbo and Lu, Xin and Yang, Jiarong and Hu, Anya and Wang, Kunyu and Xiao, Jie and Li, Dong and Fu, Xueyang and Zha, Zheng-Jun and Fan, Zihao and Wang, Xi and Zhu, Yurui and Liu, Kean and Xu, Senyan and Liu, Hongjian and Xiao, Yupeng and Serrano-Lozano, David and Molina-Bakhos, Francisco A. and Xue, Danna and Yang, Yixiong and Pilligua, Maria and Baldrich, Ramon and Vanrell, Maria and Vazquez-Corral, Javier and Sun, Xuan and Lou, Zijie and Liu, Ting and Purohit, Kuldeep and Pinjari, Jameer Babu and Zhang, Yilin and Zheng, Huan and Wei, Yanyan and Zhao, Suiyi and Tang, Shengeng and Zhang, Zhao and Zuo, Yushen and He, Zongqi and Xiao, Zhe and Yang, Cuixin and Dong, Rongkang and Xiao, Jun and Lam, Kin-Man and Akalwadi, Nikhil and Aralikatti, Vijayalaxmi Ashok and Hegde, Dheeraj Damodhar and Tabib, Ramesh Ashok and Mudenagudi, Uma and Ali, Anas M. and Benjdira, Bilel and Boulila, Wadii}, title = {NTIRE 2025 Ambient Lighting Normalization Challenge Report}, booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops}, month = {June}, year = {2025}, pages = {1289-1300} }