ITU Firearms Detection Dataset

Starting from September 2017, we at IML have collected and annotated a diverse and challenging dataset consisting of 11,000 images. This collected dataset spans over two classes of Firearms, i.e., Rifles and Guns. Axis-aligned bounding box (AABB) of each firearm in each image has been hand-annotated. Dataset has been divided into training and testing splits, where for the testing split Oriented Bounding Boxes (OBB) were also manually annotated to enable comparison with existing OBB predicting algorithms. As per our knowledge, ITUF is the first large firearm dataset in the public domain.

ITUF captures varied scenes (indoor, outdoor, lighting conditions) & scenarios (firearms pointed, carried, lying on tables/ground/racks) and contains various makes and models of firearms (from pistols to AK-47). This diversity makes ITUF a challenging and realistic dataset for the firearm detection task. Orientation-wise dataset distribution and sample images from dataset.

We believe that this dataset will help the researchers to develop algorithms for firearm detection not just for security but also for the multi-media content analysis, including AR and VR environments as well. It will also help media and content distribution companies to categorize what content is feasible for age-appropriate consumption.  The ITUF dataset is shared with multiple researchers to date and is publicly available.

Dataset (DATASET is available upon request [Google Form])

Orientation Aware Object Detection (OAOD):

Orientation aware weakly supervised object detection with application to firearms has been available on arXiv since 2019 (Link: https://arxiv.org/abs/1904.10032) and later published at Elsevier Journal of Neurocomputing (Impact Factor: 5.719) with the title “Leveraging Orientation for Weakly Supervised Object Detection with Application to Firearm Localization”. 

In this work, we propose a weakly supervised Orientation Aware Object Detection (OAOD) algorithm. Unlike existing oriented object detectors, the proposed OAOD learns to detect oriented object bounding boxes (OBB) without using OBB ground truths, i.e., by only using AxisAligned Bounding Boxes (AABB) for training. The OAOD algorithm is evaluated on the ITUF dataset and compared with the existing state-of-the-art object detectors, including fully supervised oriented object detectors. Visit the project page for more details.

Project: Orientation Aware Firearms Detection – IM (itu.edu.pk)

Localizing Firearm Carriers:

“Localizing Firearm Carriers By Identifying Human-Object Pairs” published at IEEE International Conference on Image Processing, 2020. In this work, we present a novel approach to address the problem of human-object interaction in the context of firearms carrier identification. To train and evaluate, we have extended the ITUF dataset by adding more images and annotations for Humans and Firearms bounding boxes as well as firearm-carrier information. The project page and paper links are provided below.

Project: Localizing Firearm Carriers – IM (itu.edu.pk)

Keywords:

Object Detection, Orientation Aware Object Detection, Oriented Object Detection, Firearms Detection, Orientation Aware Firearms Detection, Deep Learning and Computer Vision for object detection, Orientation Aware Object Detection with Application to Firearms