Action recognition and object detection dataset for firearm-related actions

Published: 27 November 2023| Version 2 | DOI: 10.17632/bbzpxhd22j.2


Action recognition and object detection dataset for firearm-related actions. The dataset is licensed under CC BY NC 3.0. However, we do allow commercial use, but only if prior permission is obtained.


Steps to reproduce

The dataset can be described the following way. 1. Main Dataset Folder (Gun_Action_Recognition_Dataset) : contains three subfolders corresponding to the three categories/classes: "Handgun," "Machine_Gun," and "No_Gun." 2. Category Subfolders (e.g., Handgun) : each category folder contains multiple subfolders, one for each video recorded. The names of these subfolders depend on the category and the action carried out by the subject. Within each video subfolder, you'll find the following components: - "video.mp4": the video in .mp4 format, featuring an individual performing specific actions related to the category. - "label.json": contains the details of the "video.mp4" video. about bounding boxes (xywh format), frame IDs, and object categories (obviously, the "No_Gun" category does not contain annotations information, as there are no detections), as well as general information and licenses.


Universidad de Castilla-La Mancha Escuela Tecnica Superior de Ingenieros Industriales


Computer Vision, Object Detection, Machine Learning, Pattern Recognition, Recognition, Weapon Carrying, Deep Learning, Action Recognition


European Union NextGenerationEU/PRTR

Horizon Europe dAIEdge

Grant n. 101120726

Junta de Comunidades de Castilla-La Mancha


Ministerio de Ciencia e Innovación