UAVScenes
收藏资源简介：
UAVScenes是一个大规模数据集，旨在为2D和3D模态的各种任务提供基准。该基准数据集基于经过良好校准的多模态无人机数据集MARS-LVIG构建，最初仅用于同时定位与地图构建（SLAM）。我们通过为图像和LiDAR点云提供手动标记的语义注释以及精确的6自由度（6-DoF）姿态来增强该数据集。这些新增功能支持广泛的无人机感知任务，包括检测、分割、深度估计、6-DoF定位、地点识别和新视角合成（NVS）。据我们所知，这是第一个提供图像和LiDAR点云语义注释（12万标记对）的无人机基准数据集，有望推动多模态无人机感知研究的发展。
UAVScenes is a large-scale dataset developed to establish benchmarks for diverse tasks across 2D and 3D modalities. Built upon the well-calibrated multimodal UAV dataset MARS-LVIG—originally designed exclusively for Simultaneous Localization and Mapping (SLAM)—we enhance this resource by supplementing it with manually annotated semantic labels and precise 6-degree-of-freedom (6-DoF) poses for both images and LiDAR point clouds. This expanded dataset supports a broad spectrum of UAV perception tasks, including object detection, semantic segmentation, depth estimation, 6-DoF localization, place recognition, and novel view synthesis (NVS). To the best of our knowledge, UAVScenes is the first UAV benchmark dataset offering semantic annotations for both images and LiDAR point clouds, with a total of 120,000 labeled pairs, and it is poised to advance research in multimodal UAV perception.
UAVScenes数据集概述
基本信息
- 数据集名称：UAVScenes
- 发布会议：ICCV 2025
- 论文链接：http://arxiv.org/abs/2507.22412
- 许可证：Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License（仅限学术使用）
数据集简介
- 多模态无人机数据集，支持2D和3D任务基准测试
- 基于MARS-LVIG数据集构建，增加了手动标注的语义信息
- 包含图像和LiDAR点云的语义标注（共120k标注对）
- 提供精确的6自由度（6-DoF）位姿信息
数据内容
- Hikvision相机图像及标注
- Livox Avia LiDAR点云及标注
- 6-DoF位姿数据
- 重建的3D点云/网格地图
- 颜色映射文件（cmap.py）
- 相机-LiDAR标定数据（calibration_results.py）
- 相机-3D地图标定数据（sampleinfos_interpolated.json）
数据集规格
- 包含4个大场景：
- AMtown
- AMvalley
- HKairport
- HKisland
- 每个场景包含多个运行序列（如01、02、03）
下载选项
- 完整数据集（interval=1）
- 关键帧数据集（interval=5，大小为完整数据集的1/5）
- 下载平台：
- OneDrive：https://entuedu-my.sharepoint.com/:f:/g/personal/wang1679_e_ntu_edu_sg/EgY6DU5GBchIiAIa-eQZmEAB0vJx3khCPHbFW3LnR77RFw?e=26GaSc
- Google Drive：https://drive.google.com/drive/folders/1HSJWc5qmIKLdpaS8w8pqrWch4F9MHIeN?usp=sharing
- 百度网盘：https://pan.baidu.com/s/13CgnxRFqevQ8Fa1Y3dkM0A?pwd=1679
标注工具
- 2D标注：X-AnyLabeling
- 3D标注：CloudCompare
- 3D重建：DJI Terra（大疆智图）
引用格式
bibtex @article{Wang2025UAVScenes, title = {UAVScenes: A Multi‑Modal Dataset for UAVs}, author = {Wang, Sijie and Li, Siqi and Zhang, Yawei and Yu, Shangshu and Yuan, Shenghai and She, Rui and Guo, Quanjiang and Zheng, JinXuan and Howe, Ong Kang and Chandra, Leonrich and Srijeyan, Shrivarshann and Sivadas, Aditya and Aggarwal, Toshan and Liu, Heyuan and Zhang, Hongming and Chen, Chujie and Jiang, Junyu and Xie, Lihua and Tay, Wee Peng}, journal = {arXiv preprint arXiv:2507.22412}, year = {2025}, url = {https://arxiv.org/abs/2507.22412} }
