NTIRE2026_CDFSOD
收藏资源简介:
该数据集用于NTIRE 2026挑战赛的跨域少样本目标检测任务,包含COCO作为源数据,以及ArTaxOr、Clipart1k和DeepFish作为验证数据集。数据集的组织结构包括annotations、test和train目录。
This dataset is intended for the cross-domain few-shot object detection task of the NTIRE 2026 Challenge. It uses COCO as the source dataset, with ArTaxOr, Clipart1k, and DeepFish serving as its validation datasets. The directory structure of the dataset includes annotations, test, and train folders.
NTIRE2026_CDFSOD 数据集概述
数据集基本信息
- 数据集名称:NTIRE2026_CDFSOD
- 关联挑战:NTIRE 2026 Challenge on 2-nd Cross-Domain Few-Shot Object Detection @ CVPR 2026
- 官方链接:https://www.codabench.org/competitions/12873/
- 挑战平台:https://codalab.lisn.upsaclay.fr/competitions/21851
挑战任务
- 任务类型:跨域少样本目标检测
- 核心目标:在跨域场景下,使用极少量的带标签目标图像,提升目标检测模型的性能。
- 评估指标:平均精度均值
- 数据使用限制:不限制源数据选择或预训练模型的使用,但禁止手动搜索更多的支持图像。
数据集构成
源数据
- 采用COCO数据集作为源数据。
验证数据集
- ArTaxOr
- Clipart1k
- DeepFish
数据下载与组织
-
下载地址:https://drive.google.com/drive/folders/16SDv_V7RDjTKDk8uodL2ubyubYTMdd5q?usp=drive_link
-
目录结构:
|NTIRE2026_CDFSOD/datasets/ |--clipart1k/ | |--annotations | |--test | |--train |--ArTaxOr/ | |--annotations | |--test | |--train |--......
-
权重文件组织:
|NTIRE2026_CDFSOD/weights/ |--trained/ | |--vitl_0089999.pth |--background/ | |--background_prototypes.vitl14.pth
基线模型与资源
提供的基线模型
- DE-ViT
- CD-ViTO
推荐的解决方案
- ETS:第一届公开赛道获奖方案。GitHub地址:https://github.com/jaychempan/ETS
- Domain-RAG:当前最先进方法。GitHub地址:https://github.com/LiYu0524/Domain-RAG
- X-Few:第一届闭源赛道获奖方案。GitHub地址:https://github.com/johnmaijer/X-Few-_CD-FSOD
其他相关解决方案链接
- https://github.com/KAIJINZ228/Few_Shot_GD
- https://github.com/Pumpkinder/GLIP-CDFSOD
- https://github.com/omnipotent13/cfsod_challenge
- https://github.com/lch216/cdfsod-project
- https://github.com/Sun-HR02/ProtoDINO/tree/main
- https://github.com/Weston111/cdfsod
- https://github.com/yuqing-yang/CDFSOD-FSV
- https://github.com/Linzeyin/NTIRE2025_CDFSOD_tta
评估与提交
- 提交格式:必须使用COCO风格的JSON标注格式提交预测结果。
- 复现声明:每次提交必须包含复现声明,详细说明模型的训练策略和使用的预训练资源。
- 排名与奖励:排名靠前的参与者将获得奖励,并被邀请按照CVPR研讨会提交指南描述其解决方案,并提交至CVPR 2026相关的NTIRE研讨会。
组织者联系方式
- Xingyu Qiu (xyqiu24@m.fudan.edu.cn)
- Yuqian Fu (yuqian.fu.ai@gmail.com)
- Jiawei Geng (jwgeng25@m.fudan.edu.cn)
- Bin Ren (bin.ren.mondo@gmail.com)
- Jiancheng Pan (jiancheng.pan.plus@gmail.com)
- Zongwei Wu (zongwei.wu@uni-wuerzburg.de)
- Hao Tang (howard.haotang@gmail.com)
- Yanwei Fu (yanweifu@fudan.edu.cn)
- Radu Timofte (radu.timofte@uni-wuerzburg.de)
- Nicu Sebe (niculae.sebe@unitn.it)
- Mohamed H. Elhoseiny (mohamed.elhoseiny@kaust.edu.sa)
参考文献
- Fu, Y., Wang, Y., Pan, Y., Huai, L., Qiu, X., Shangguan, Z., Liu, T., Fu, Y., Van Gool, L., & Jiang, X. (2025). Cross-domain few-shot object detection via enhanced open-set object detector. In European Conference on Computer Vision (pp. 247–264). Springer.
- Qiu, X., Fu, Y., Geng, J., Ren, B., Pan, J., Fu, Y., Timofte, R., & others. (2026). NTIRE 2026 challenge on cross-domain few-shot object detection: methods and results. In CVPRW.
- Li, Y., Qiu, X., Fu, Y., Chen, J., Qian, T., Zheng, X., Paudel, D. P., Fu, Y., Huang, X., Van Gool, L., & others. (2025). Domain-RAG: Retrieval-Guided Compositional Image Generation for Cross-Domain Few-Shot Object Detection. arXiv preprint arXiv:2506.05872.
- Fu, Y., Xie, Y., Fu, Y., & Jiang, Y.-G. (2023). Styleadv: Meta style adversarial training for cross-domain few-shot learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 24575–24584).
- Fu, Y., Fu, Y., & Jiang, Y.-G. (2021). Meta-fdmixup: Cross-domain few-shot learning guided by labeled target data. In Proceedings of the 29th ACM international conference on multimedia (pp. 5326–5334).




