UAVScenes is a large-scale dataset introduced for ICCV 2025 to benchmark tasks across 2D and 3D modalities. It is built upon the MARS-LVIG dataset and enhanced with manually labeled semantic annotations for images and LiDAR point clouds. The dataset was uploaded to Hugging Face by user sijieaaa on August 6, 2025.
Use Cases
- Benchmarking 2D semantic segmentation models based on manually labeled image annotations
- Developing 3D perception algorithms for UAVs based on annotated LiDAR point clouds
- Training multi-modal fusion models for UAV navigation using the combined image and point cloud data
- Evaluating simultaneous localization and mapping (SLAM) performance in outdoor environments
Strengths
- Dataset is described as 'large-scale' in the description
- Provides manually labeled semantic annotations for both images and LiDAR point clouds
- Built upon a well-calibrated multi-modal UAV dataset (MARS-LVIG)
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is unknown, which may limit suitability assessment
Provenance
- Source
- sijieaaa on Hugging Face, based on the MARS-LVIG dataset
- Collection Method
- Enhanced from an existing UAV dataset with manual semantic labeling
- Time Range
- null
- Freshness
- Last updated 2025-08 06 08:49:56; freshness should be verified
- Geography
- null