NTIRE 2026 视频注意力预测挑战：方法与结果概述
计算机视觉与模式识别
2026-04-17 v1 人机交互
多媒体
摘要
本文概述了 NTIRE 2026 视频注意力预测挑战。挑战参赛者的目标是开发用于预测提供视频序列中注意力图的自动方法。为本挑战准备了一套新数据集，包含 2000 个多样化视频并采用开放许可证。使用众包鼠标跟踪收集的 fixations 以及对应的注意力图，包含来自超过 5000 名评估者的观看数据。使用常接受的质量指标对 800 个测试视频子集上的 80 个团队进行评估，7 个团队通过最终阶段的代码审查。本挑战中使用的所有数据均公开可用 - https://github.com/msu-video-group/NTIRE26_Saliency_Prediction。
引用
@article{arxiv.2604.14816,
title = {NTIRE 2026 Challenge on Video Saliency Prediction: Methods and Results},
author = {Andrey Moskalenko and Alexey Bryncev and Ivan Kosmynin and Kira Shilovskaya and Mikhail Erofeev and Dmitry Vatolin and Radu Timofte and Kun Wang and Yupeng Hu and Zhiran Li and Hao Liu and Qianlong Xiang and Liqiang Nie and Konstantinos Chaldaiopoulos and Niki Efthymiou and Athanasia Zlatintsi and Panagiotis Filntisis and Katerina Pastra and Petros Maragos and Li Yang and Gen Zhan and Yiting Liao and Yabin Zhang and Yuxin Liu and Xu Wu and Yunheng Zheng and Linze Li and Kun He and Cong Wu and Xuefeng Zhu and Tianyang Xu and Xiaojun Wu and Wenzhuo Zhao and Keren Fu and Gongyang Li and Shixiang Shi and Jianlin Chen and Haibin Ling and Yaoxin Jiang and Guoyi Xu and Jiajia Liu and Yaokun Shi and Jiachen Tu},
journal= {arXiv preprint arXiv:2604.14816},
year = {2026}
}
备注
CVPRW 2026