Weapon Detection: Handgun vs Bag of Chips

In 2024, a student was handcuffed after a weapon detection AI flagged a bag of Doritos as a handgun. This dataset was created to solve that exact problem. Using Hard Negative Mining, it contains a balanced split of images showing people holding handguns (real threats) versus people holding shiny, crinkly chip bags (common false positive triggers). Both objects share similar visual properties from CCTV angles: compact shape, metallic/reflective surfaces, and single-hand grip. The dataset forces models to learn the critical differences between threatening and harmless objects. Images are rendered from realistic overhead and angled CCTV perspectives with varied lighting, sensor noise, and motion blur. Annotated in YOLO format with classes for person, handgun, and chip bag. The open-source sample includes 110 images; the full package contains 1,000+ images with evaluation videos.