Research map / Detection and segmentation / CNN features
M$^2$E-UAV: A Benchmark and Analysis for Onboard Motion-on-Motion Event-Based Tiny UAV Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces M²E-UAV, an onboard event-IMU dataset and benchmark for detecting tiny UAVs when both the event camera carrier and target UAV move, and benchmarks event-frame, voxel-grid, and point-set baselines.
Problem
Tiny UAV detection from an onboard event camera when both observer and target move (motion-on-motion), where ego-motion produces dense background events and the target is a sparse event cluster.
Approach
A dataset of synchronized 720p event streams and IMU from a UAV-mounted sensor observing a moving tiny UAV, with 10 Hz box annotations temporally propagated into point-level/event-level foreground labels, plus a unified benchmark across event representations.
Key results
- RVT (voxel grid) achieves the best reported results with F1 0.8643, mAP50 0.8186 and mAP50:95 0.3815 on the validation split.
- YOLOv10 on event frames reaches F1 0.7283 and mAP50 0.6349, with mAP50:95 0.2566.
- RandLA-Net on point sets gets mAP50 0.0228 and mAP50:95 0.0117, and EV-UAV gets F1 0.0863, showing point-set transfers are weak.
Surprising findings
- The EV-UAV point-set baseline has high mAP50 (0.3761) yet very low F1 (0.0863), and RandLA-Net has mAP50 of only 0.0228.
Limitations
- Only four illumination-background scene families; no weather, altitude, distance, or target-motion variation.
- Results are reported on the validation split; held-out test evaluation is deferred to a separate release.
- Baselines are run with official default configurations, not tuned for this benchmark.
Experimental setup
Datasets: M²E-UAV
Benchmarks: M²E-UAV validation split
Metrics: F1 mAP50 mAP50:95
Baselines: YOLOv10 RVT SAST EV-UAV KPConv RandLA-Net COSeg
Builds on: RVT SAST EV-UAV KPConv RandLA-Net
Closest papers by meaning
- UAV3D: A Large-scale 3D Perception Benchmark for Unmanned Aerial Vehicles (NeurIPS 2024 · similarity 0.84)
- RemDet: Rethinking Efficient Model Design for UAV Object Detection (AAAI 2025 · similarity 0.84)
- PointSR: Self-Regularized Point Supervision for Drone-View Object Detection (CVPR 2025 · similarity 0.84)
- MMOT: The First Challenging Benchmark for Drone-based Multispectral Multi-Object Tracking (NeurIPS 2025 · similarity 0.84)
- MM-Tracker: Motion Mamba for UAV-platform Multiple Object Tracking (AAAI 2025 · similarity 0.83)
- ETAP: Event-based Tracking of Any Point (CVPR 2025 · similarity 0.83)
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.