VQualA 2025 挑战：面部图像质量评估的方法与结果
计算机视觉与模式识别
2025-08-27 v1
摘要
人脸图像在众多应用中发挥着关键作用；然而，现实世界条件常常会引入噪声、模糊和压缩伪影等退化，影响整体图像质量并阻碍后续任务。为解决这一挑战，我们组织了 VQualA 2025 挑战，旨在进行面部图像质量评估（FIQA），作为 ICCV 2025 工作坊的一部分。参赛者创建了轻量级高效模型（限制为0.5 GFLOPs和500万参数），用于预测来自任意分辨率和真实退化人脸图像的平均意见分数（MOS）。提交的方案通过来自野生人脸图像数据集上的相关性指标进行全面评估。该挑战吸引了127位参赛者，最终提交了1519个提交方案。本报告总结了推动实用 FIQA 方法发展的方法与发现。
引用
@article{arxiv.2508.18445,
title = {VQualA 2025 Challenge on Face Image Quality Assessment: Methods and Results},
author = {Sizhuo Ma and Wei-Ting Chen and Qiang Gao and Jian Wang and Chris Wei Zhou and Wei Sun and Weixia Zhang and Linhan Cao and Jun Jia and Xiangyang Zhu and Dandan Zhu and Xiongkuo Min and Guangtao Zhai and Baoying Chen and Xiongwei Xiao and Jishen Zeng and Wei Wu and Tiexuan Lou and Yuchen Tan and Chunyi Song and Zhiwei Xu and MohammadAli Hamidi and Hadi Amirpour and Mingyin Bai and Jiawang Du and Zhenyu Jiang and Zilong Lu and Ziguan Cui and Zongliang Gan and Xinpeng Li and Shiqi Jiang and Chenhui Li and Changbo Wang and Weijun Yuan and Zhan Li and Yihang Chen and Yifan Deng and Ruting Deng and Zhanglu Chen and Boyang Yao and Shuling Zheng and Feng Zhang and Zhiheng Fu and Abhishek Joshi and Aman Agarwal and Rakhil Immidisetti and Ajay Narasimha Mopidevi and Vishwajeet Shukla and Hao Yang and Ruikun Zhang and Liyuan Pan and Kaixin Deng and Hang Ouyang and Fan yang and Zhizun Luo and Zhuohang Shi and Songning Lai and Weilin Ruan and Yutao Yue},
journal= {arXiv preprint arXiv:2508.18445},
year = {2025}
}
备注
ICCV 2025 VQualA workshop FIQA track