Research map / Point clouds / Segmentation and detection
UAVScenes: A Multi-Modal Dataset for UAVs
Sijie Wang, Siqi Li, Yawei Zhang, Shangshu Yu, Shenghai Yuan, Rui She, Quanjiang Guo, JinXuan Zheng, Ong Kang Howe, Leonrich Chandra, Shrivarshann Srijeyan, Aditya Sivadas and 7 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use a Transformer backbone (DeiT3-m) instead of CNN (ResNet-50) | UperNet segmentation · UAVScenes image semantic segmentation | mIoU: 61.3 -> 68.3 | Improved |
| use ViT-t instead of ConvNext-t | UperNet segmentation · UAVScenes image semantic segmentation | mIoU: 55.3 -> 62.8 | Improved |
| use PTv2 (11M) instead of MinkUNet (38M) | LiDAR segmentation · UAVScenes LiDAR semantic segmentation | mIoU: 32.7 -> 33.2 | Improved |
| use fusion (MinkLoc++) instead of LiDAR only (MinkLoc3D V2) | place recognition · UAVScenes place recognition | Recall@1: 42.8 -> 47.1 | Improved |
| use DINOv2 foundation backbone (SALAD) instead of ResNet-18 (GeM) | camera place recognition · UAVScenes place recognition | Recall@1: 42.1 -> 67.1 | Improved |
| use BEV projection (BEVPlace) instead of voxels (MinkLoc3D) | LiDAR place recognition · UAVScenes place recognition | Recall@1: 41.9 -> 32.6 | Hurt |
| use fusion UMF instead of single-modal MinkLoc3D | place recognition · UAVScenes place recognition | Recall@1: 41.9 -> 40.1 | No clear effect |
| use absolute pose regression (RobustLoc) instead of scene coordinate regression (ACE) | 6-DoF localization · UAVScenes localization average | median position error (m): 132.9 -> 6.1 | Improved |
| use 3DGS instead of Instant-NGP | novel view synthesis · UAVScenes NVS AMtown | PSNR: 20.09 -> 22.95 | Improved |
| use Metric3D V2 (ViT-l) instead of Depth Anything V2 (ViT-l) | zero-shot depth · UAVScenes depth | AbsRel: 1.517 -> 0.540 | Improved |
| use single-step models instead of diffusion models (Marigold) | zero-shot depth · UAVScenes depth | AbsRel: 0.994 -> 0.540 | Improved |
Closest papers by meaning
- UAV3D: A Large-scale 3D Perception Benchmark for Unmanned Aerial Vehicles (NeurIPS 2024 · similarity 0.90)
- Is your VLM Sky-Ready? A Comprehensive Spatial Intelligence Benchmark for UAV Navigation (CVPR 2026 · similarity 0.88)
- SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery (NeurIPS 2026 · similarity 0.88)
- AGC-Drive: A Large-Scale Dataset for Real-World Aerial-Ground Collaboration in Driving Scenarios (NeurIPS 2025 · similarity 0.88)
- Towards Realistic UAV Vision-Language Navigation: Platform, Benchmark, and Methodology (ICLR 2025 · similarity 0.88)
- Multiview Aerial Visual RECognition (MAVREC): Can Multi-view Improve Aerial Visual Perception? (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
