Vigil AI
One software stack for counter-UAS perception: mission-grade training data, independent performance validation, and deployment-ready models built for contested airspace.
Vigil Dataset
The foundational training dataset for counter-UAS computer vision. 500K+ annotations with 1 cm RTK positioning, coordinated timestamps, and camera intrinsics matched to your platform.
Vigil Benchmark
Independent C-UAS model evaluation. One rigorous certification standard for counter-UAS detection readiness.
Vigil Model
Pre-trained counter-UAS perception. State-of-the-art detection, tracking, and segmentation across visible and infrared.
Vigil Dataset
The foundational dataset for counter-UAS computer vision. CommonDefense delivers densely annotated imagery of UAS targets across every operational scenario — from Group 1 micro-drones against cluttered urban backgrounds to fixed-wing platforms at range against open sky.
80+ C-UAS-relevant object classes. Pixel-level instance segmentation. Oriented bounding boxes. Multi-spectral coverage across RGB, LWIR, and MWIR — because real-world counter-drone operations don't happen in a single spectrum.
Every annotation ships with RTK positioning accurate to 1 cm, coordinated timestamps across all trajectories, and full camera calibration intrinsics — tunable to match your platform. No guesswork. Plug it straight into your pipeline.
Built for teams shipping counter-UAS systems, not publishing papers.
4096 × 3072 · RGB · GSD 0.3m
Vigil Benchmark
| System | Detection mAP | Tracking MOTA | Classification F1 | Latency P95 | Status |
|---|---|---|---|---|---|
| System A | 91.2% | 82.1% | 0.89 | 87ms | Certified |
| System B | 86.4% | 76.3% | 0.84 | 112ms | Certified |
| System C | 78.9% | 68.7% | 0.79 | 156ms | Certified |
| System D | 72.1% | 61.2% | 0.71 | 203ms | Not Certified |
| Your System | — | — | — | — | ? |
Vigil Model
C-UAS Detection Performance
Off-the-shelf C-UAS perception. BaseModel ships pre-trained on CommonDefense — delivering state-of-the-art drone detection, tracking, and instance segmentation across visible and infrared. No training pipeline. No ML team. Deploy to your counter-drone system this week.
Optimized for real-time inference on NVIDIA Jetson, GPU servers, and ONNX-compatible runtimes. Handles fixed-wing UAS, rotary-wing UAS, and Group 1–3 platforms across cluttered backgrounds, adverse weather, and low-contrast thermal scenes.
Trained on C-UAS data. Validated on C-UAS scenarios. Deployed by C-UAS operators.