UAVDB: Point-Guided Masks for UAV Detection and Segmentation
This work addresses the need for accurate and scalable UAV detection solutions in surveillance and security, though it is incremental as it builds upon existing datasets and models like SAM2.
The authors tackled the lack of large-scale, high-resolution datasets for UAV detection and segmentation by introducing UAVDB, a benchmark dataset built with a point-guided weak supervision pipeline, which achieved superior IoU accuracy and annotation efficiency compared to existing techniques.
The widespread deployment of Unmanned Aerial Vehicles (UAVs) in surveillance, security, and airspace monitoring demands accurate and scalable detection solutions. However, progress is hindered by the lack of large-scale, high-resolution datasets with precise and cost-effective annotations. We present UAVDB, a new benchmark dataset for UAV detection and segmentation, built upon a point-guided weak supervision pipeline. As its foundation, UAVDB leverages trajectory point annotations and RGB video frames from the multi-view drone tracking dataset, captured by fixed-camera setups. We introduce an efficient annotation method, Patch Intensity Convergence (PIC), which generates high-fidelity bounding boxes directly from these trajectory points, eliminating manual labeling while maintaining accurate spatial localization. We further derive instance segmentation masks from these bounding boxes using the second version of the Segment Anything Model (SAM2), enabling rich multi-task annotations with minimal supervision. UAVDB captures UAVs at diverse scales, from visible objects to near-single-pixel instances, under challenging environmental conditions. Particularly, PIC is lightweight and readily pluggable into other point-guided scenarios, making it easy to scale up dataset generation across domains. We quantitatively compare PIC against existing annotation techniques, demonstrating superior Intersection over Union (IoU) accuracy and annotation efficiency. Finally, we benchmark several state-of-the-art (SOTA) YOLO-series detectors on UAVDB, establishing strong baselines for future research. The source code is available at https://github.com/wish44165/UAVDB .