VQualA 2025 图像超分辨率生成内容质量评估挑战：方法与结果
计算机视觉与模式识别
2025-09-09 v1 图像与视频处理
摘要
本文介绍了 ISRGC-Q 挑战，基于 Image Super-Resolution Generated Content Quality Assessment (ISRGen-QA) 数据集构建，并作为 ICCV 2025 研讨会的 Visual Quality Assessment (VQualA) 比赛组织。不同于现有的超分辨率图像质量评估 (SR-IQA) 数据集，ISRGen-QA 更强调由最新生成方法（包括生成对抗网络 (GAN) 和扩散模型）生成的超分辨率图像。该挑战的主要目标是分析现代超分辨率技术引入的独特瑕疵，并有效评估其感知质量。本次挑战共有 108 名参赛者报名，4 支队伍提交了有效解答和测试阶段的事业表。这些提交在 ISRGen-QA 数据集上展示了最先进 (SOTA) 的性能。该项目已公开发布于：https://github.com/Lighting-YXLI/ISRGen-QA。
引用
@article{arxiv.2509.06413,
title = {VQualA 2025 Challenge on Image Super-Resolution Generated Content Quality Assessment: Methods and Results},
author = {Yixiao Li and Xin Li and Chris Wei Zhou and Shuo Xing and Hadi Amirpour and Xiaoshuai Hao and Guanghui Yue and Baoquan Zhao and Weide Liu and Xiaoyuan Yang and Zhengzhong Tu and Xinyu Li and Chuanbiao Song and Chenqi Zhang and Jun Lan and Huijia Zhu and Weiqiang Wang and Xiaoyan Sun and Shishun Tian and Dongyang Yan and Weixia Zhang and Junlin Chen and Wei Sun and Zhihua Wang and Zhuohang Shi and Zhizun Luo and Hang Ouyang and Tianxin Xiao and Fan Yang and Zhaowang Wu and Kaixin Deng},
journal= {arXiv preprint arXiv:2509.06413},
year = {2025}
}
备注
11 pages, 12 figures, VQualA ICCV Workshop