NTIRE 2026 真实世界人脸恢复第二挑战：方法与结果综述
计算机视觉与模式识别
2026-04-16 v2
摘要
本文综述了 NTIRE 2026 挑战关于真实世界人脸恢复的提出方案和所得结果。该挑战聚焦于生成自然真实的输出，同时保持身份一致性。其目标是推动感知质量和真实性的前沿解决方案，同时不限制计算资源或训练数据。性能采用加权图像质量评估 (IQA) 分数评估，并使用 AdaFace 模型作为身份检查器。本次竞赛吸引了 96 名报名者，其中 10 支队伍提交了有效模型；最终有 9 支队伍在最终排名中取得有效分数。该协作努力推进了真实世界人脸恢复的性能，同时提供了该领域最新趋势的深入概述。
引用
@article{arxiv.2604.10532,
title = {The Second Challenge on Real-World Face Restoration at NTIRE 2026: Methods and Results},
author = {Jingkai Wang and Jue Gong and Zheng Chen and Kai Liu and Jiatong Li and Yulun Zhang and Radu Timofte and Jiachen Tu and Yaokun Shi and Guoyi Xu and Yaoxin Jiang and Jiajia Liu and Yingsi Chen and Yijiao Liu and Hui Li and Yu Wang and Congchao Zhu and Alexandru-Gabriel Lefterache and Anamaria Radoi and Chuanyue Yan and Tao Lu and Yanduo Zhang and Kanghui Zhao and Jiaming Wang and Yuqi Li and WenBo Xiong and Yifei Chen and Xian Hu and Wei Deng and Daiguo Zhou and Sujith Roy and Claudia Jesuraj and Vikas B and Spoorthi LC and Nikhil Akalwadi and Ramesh Ashok Tabib and Uma Mudenagudi and Yuxuan Jiang and Chengxi Zeng and Tianhao Peng and Fan Zhang and David Bull Wei Zhou and Linfeng Li and Hongyu Huang and Hoyoung Lee and SangYun Oh and ChangYoung Jeong and Axi Niu and Jinyang Zhang and Zhenguo Wu and Senyan Qing and Jinqiu Sun and Yanning Zhang},
journal= {arXiv preprint arXiv:2604.10532},
year = {2026}
}
备注
NTIRE 26: https://cvlai.net/ntire/2026 . NTIRE Real-World Face Restoration: https://ntire-face.github.io/2026/ . CVPR 2026 Workshop