Synthetic CCTV Weapon Detection Image Data | Handgun, Rifle, Shotgun | YOLO + COCO Annotations

Weapon detection models fail in the real world for two reasons: too few training images of guns seen from real CCTV angles, and too many false alarms on everyday objects. This dataset tackles both. What's inside - Synthetic CCTV images of people carrying or holding firearms in shops, offices, stations, warehouses, petrol forecourts and streets. - Ceiling, wall and pole-mounted camera views, by day, at night and in monochrome IR mode. - Weapons appear small and distant as well as close, the way they do on real security cameras. - About a fifth of the images are hard negatives (bottles, umbrellas, bags, phones, tools) so the model learns what not to flag. - Classes: person, handgun, rifle, shotgun, sniper_rifle. - Every person and weapon boxed, in YOLO and COCO format, with a train/val split, data.yaml and a dataset card. Volumes - 1,000-image pack available off the shelf (zip download). - Up to 20,000 images available on request, plus 100+ matching CCTV video clips (720p-1080p, 20 FPS, MP4). - Custom packs for your camera angles, sites and weapon types. Why synthetic - No real people, faces or locations, so no GDPR or consent issues. - Rare and dangerous scenarios at volume, without staging them. - Every image reviewed for realism and label accuracy before release. Use cases: weapon detection for VMS and video analytics, school and retail security, transport hubs, model benchmarking and false-positive reduction. Free samples are on Kaggle and Hugging Face. The provider (Swegardt