NTIRE 2025 挑战：真实人脸恢复方法与结果综述
计算机视觉与模式识别
2025-04-22 v1
摘要
本文提供了 NTIRE 2025 挑战中关于真实人脸恢复的综述，突出提出的解决方案及其结果。该挑战聚焦于生成自然、逼真的输出，同时保持身份一致性。其目标是推动感知质量和真实性的 SOTA 解决方案发展，不限制计算资源或训练数据。挑战的评估轨道采用加权图像质量评估(IQA)得分，并使用 AdaFace 模型作为身份检查器。本次竞赛吸引了 141 名报名者，其中 13 支队伍提交了有效模型，最终有 10 支队伍在最终排名中取得有效得分。该协作努力推进了真实人脸恢复的性能，同时提供了该领域最新趋势的深入概述。
引用
@article{arxiv.2504.14600,
title = {NTIRE 2025 Challenge on Real-World Face Restoration: Methods and Results},
author = {Zheng Chen and Jingkai Wang and Kai Liu and Jue Gong and Lei Sun and Zongwei Wu and Radu Timofte and Yulun Zhang and Jianxing Zhang and Jinlong Wu and Jun Wang and Zheng Xie and Hakjae Jeon and Suejin Han and Hyung-Ju Chun and Hyunhee Park and Zhicun Yin and Junjie Chen and Ming Liu and Xiaoming Li and Chao Zhou and Wangmeng Zuo and Weixia Zhang and Dingquan Li and Kede Ma and Yun Zhang and Zhuofan Zheng and Yuyue Liu and Shizhen Tang and Zihao Zhang and Yi Ning and Hao Jiang and Wenjie An and Kangmeng Yu and Chenyang Wang and Kui Jiang and Xianming Liu and Junjun Jiang and Yingfu Zhang and Gang He and Siqi Wang and Kepeng Xu and Zhenyang Liu and Changxin Zhou and Shanlan Shen and Yubo Duan and Yiang Chen and Jin Guo and Mengru Yang and Jen-Wei Lee and Chia-Ming Lee and Chih-Chung Hsu and Hu Peng and Chunming He},
journal= {arXiv preprint arXiv:2504.14600},
year = {2025}
}
备注
NTIRE 2025 webpage: https://www.cvlai.net/ntire/2025. Code: https://github.com/zhengchen1999/NTIRE2025_RealWorld_Face_Restoration