NTIRE 2026 跨域少样本目标检测第二挑战：方法与结果
计算机视觉与模式识别
2026-04-15 v1 人工智能
摘要
跨域少样本目标检测（CD-FSOD）仍是现有目标检测器和少样本学习方法的挑战，尤其是在不同域之间泛化时。作为NTIRE 2026的一部分，我们举办了第二届CD-FSOD挑战，以系统评估和推动在受限标注条件下检测未见目标域中目标的进展。该挑战获得了强烈的社区关注，共有128个注册参赛者，总计696篇提交作品。其中31支队伍积极参赛，19支队伍提交了有效的最终结果。参赛者探索了广泛的策略，提出创新方法，推动了两种开源和封闭源轨道下的性能前沿。本报告详细概述了NTIRE 2026 CD-FSOD挑战，包括所提交方法的概述以及所有参赛队最终结果的分析。挑战代码：https://github.com/ohMargin/NTIRE2026_CDFSOD
引用
@article{arxiv.2604.11998,
title = {The Second Challenge on Cross-Domain Few-Shot Object Detection at NTIRE 2026: Methods and Results},
author = {Xingyu Qiu and Yuqian Fu and Jiawei Geng and Bin Ren and Jiancheng Pan and Zongwei Wu and Hao Tang and Yanwei Fu and Radu Timofte and Nicu Sebe and Mohamed Elhoseiny and Lingyi Hong and Mingxi Cheng and Xingqi He and Runze Li and Xingdong Sheng and Wenqiang Zhang and Jiacong Liu and Shu Luo and Yikai Qin and Yaze Zhao and Yongwei Jiang and Yixiong Zou and Zhe Zhang and Yang Yang and Kaiyu Li and Bowen Fu and Zixuan Jiang and Ke Li and Hui Qiao and Xiangyong Cao and Xuanlong Yu and Youyang Sha and Longfei Liu and Di Yang and Xi Shen and Kyeongryeol Go and Taewoong Jang and Saiprasad Meesiyawar and Ravi Kirasur and Rakshita Kulkarni and Bhoomi Deshpande and Harsh Patil and Uma Mudenagudi and Shuming Hu and Chao Chen and Tao Wang and Wei Zhou and Qi Xu and Zhenzhao Xing and Dandan Zhao and Hanzhe Xia and Dongdong Lu and Zhe Zhang and Jingru Wang and Guangwei Huang and Jiachen Tu and Yaokun Shi and Guoyi Xu and Yaoxin Jiang and Jiajia Liu and Liwei Zhou and Bei Dou and Tao Wu and Zekang Fan and Junjie Liu and Adhémar de Senneville and Flavien Armangeon and Mengbers and Yazhe Lyu and Zhimeng Xin and Zijian Zhuang and Hongchun Zhu and Li Wang},
journal= {arXiv preprint arXiv:2604.11998},
year = {2026}
}
备注
accepted by CVPRW 26 @ NTIRE