ATR-UMMIR
收藏资源简介:
ATR-UMMIR是一个专为无人机多模态图像配准设计的基准数据集,包含7,969个图像三元组,包括高分辨率原始可见光图像、原始红外图像和精确配准的可见光图像。数据集涵盖了从80米到300米的飞行高度、0°到75°的相机角度,以及全年全天候的天气和光照条件。ATR-UMMIR为每个三元组引入了可靠的像素级真实值,并标注了六个成像条件属性,以评估配准的鲁棒性。此外,数据集还提供了对象级注释,涵盖了11个对象类别,总计有77,753个可见光和78,409个红外边框。ATR-UMMIR旨在推进无人机场景下的多模态配准、融合和感知研究。
ATR-UMMIR is a benchmark dataset specifically designed for multimodal image registration in unmanned aerial vehicle (UAV) scenarios. It consists of 7,969 image triplets, including high-resolution raw visible light images, raw infrared images, and precisely registered visible light images. The dataset covers flight altitudes ranging from 80 meters to 300 meters, camera angles from 0° to 75°, and all-weather and illumination conditions throughout the year. ATR-UMMIR provides reliable pixel-level ground truth for each triplet, and annotates six imaging condition attributes to evaluate the robustness of registration. In addition, the dataset also offers object-level annotations covering 11 object categories, with a total of 77,753 visible light and 78,409 infrared bounding boxes. ATR-UMMIR aims to advance research on multimodal registration, fusion and perception in UAV scenarios.
ATR-UMMIR 数据集概述
数据集简介
ATR-UMMIR(无人多模态图像配准)是一个用于航空场景下多模态图像配准和匹配的大规模数据集,专注于对齐无人机在不同真实世界条件下捕获的可见光和红外图像对。
数据集详情
- 模态:对齐的可见光-红外图像对
- 场景数量:15,000+ 个独特位置
- 总图像数量:60,000+(30k 可见光,30k 红外)
- 分辨率:640×512 像素
标注信息
- 粗粒度手动对齐
- 细粒度关键点(针对子集)
- 详细条件标签:
- 高度:80m–300m(主要在 100–120m)
- 相机角度:0°(天底)到 75°(倾斜),主要在 30°–45°
- 拍摄时间:白天、夜晚、黎明、早晨、下午
- 天气:晴天、多云、雨天、雨后、雾天
- 光照:夜晚、黄昏、昏暗、正常、过曝
- 场景:11 种类型,包括城市、郊区、村庄、工厂、道路、学校等
应用领域
- 多模态图像配准和对齐
- 条件感知图像匹配
- 跨模态融合和检测
- 基于无人机的遥感任务
下载信息
引用
bibtex @misc{ATRUMMIR2025, title={ATR-UMMIR: A Multimodal UAV Image Matching Dataset under Diverse Conditions}, author={Your Name and Others}, year={2025}, howpublished={url{https://github.com/yourname/ATR-UMMIR}}, }




