Synthetic CCTV and security datasets for computer vision. Weapons, knives, falls, fights, shoplifting, fire and drones, seen from real CCTV camera angles, with hard negatives, labelled in YOLO and COCO. No real people, so no privacy problem.
Simuletic is a Swedish company (since 2023) that sells synthetic CCTV and security datasets for computer vision: 16 datasets with YOLO and COCO labels, one-time licences from $149 (research) and $590 (commercial), free samples on Hugging Face, Kaggle and GitHub, and detectors measured on real footage.
Website · All datasets · Hugging Face · contact@simuletic.com
Each repo has a free sample with YOLO and COCO labels, hard negatives, a Colab notebook that fine-tunes YOLO11, and a link to the full 1,000-image dataset.
 CCTV Weapon Detection person, handgun, rifle, shotgun, sniper_rifle Full 1K dataset |
 CCTV Knife Detection person, knife Full 1K dataset |
 CCTV Fall Detection person, fallen_person Full 1K dataset |
Also: cctv-aggressive-poses-fight-detection-dataset (aggressive poses and fights) · full dataset
Full datasets: one-time licence, $149 research or $590 commercial. Larger sets (5,000 or 10,000 images) on request.
| Area | Datasets |
|---|---|
| Security CCTV | Weapon · Knife · Rifles vs umbrellas · Handgun vs bag of chips · ATM robbery · Aggressive poses / fights · Pedestrian attributes |
| Safety | Fall / lying down · Smoke and fire · Long-distance wildfire smoke · Driver drowsiness |
| Drones and aerial | Airport drone threats · Drone swarms · UAV view: battle tanks |
| Vision-language | Surveillance VLM: weapon and knife captions |
We train small detectors only on our synthetic data and score them on public real-footage test sets the models never saw.
- Fall detection model (D-FINE Small and Nano): a synthetic-only model catches 97% of people lying on the floor at 5% false alarms on CAUCAFall, and 83–87% of falls in a second home (GMDCSA-24) where a model trained on real footage from one room catches 24%. Model page · Case study with the full table · Method guide
- simuletic-vision: Python toolkit to validate YOLO-format datasets and train, evaluate and run RF-DETR detectors on synthetic and real data.
- Custom datasets rendered into your own camera views and classes, as images or video clips.
- False-Positive Stress Test: we run your detector against hard negatives and edge cases and send a failure report.