Dangerous Items Dataset for 5-Class Object Detection (YOLO annotation)

This dataset was built to carry out the research described in the paper titled Effectiveness of modern models belonging to the YOLO and Vision Transformer architectures in dangerous items detection. It contains images belonging to the following categories: objects clearly visible, objects partially covered, small objects, poor objects' sharpness, poor objects' illumination and background images. The full dataset consists of 8478 images containing a total of 8805 instances of dangerous items: machete, knife, baseball bat, rifle, gun and 826 background images (https://drive.google.com/file/d/1qDpdeou_rXDX15AryWp5Zkr6jceX_S3U/view?usp=drive_link). The dataset was randomly divided into a training, validation and testing subset (70%, 15% and 15% of the full set, respectively). The subsets are balanced, that is, they contain a similar number of instances of dangerous items. The images were annotated using bounding boxes in the YOLO format. The dataset can be used in experiments on dangerous object detection using YOLO networks.

Using this dataset please cite:
Omiotek, Z. (2025). Effectiveness of Modern Models Belonging to the YOLO and Vision Transformer Architectures in Dangerous Items Detection. Electronics, 14(17), 3540. https://doi.org/10.3390/electronics14173540