Kids must be safe at school. AI For Mankind wants to build and deploy gun detection system to local schools to protect our children.
- YouTube-GDD: A challenging gun detection dataset with rich contextual information.
- Hands Guns and Phones (HGP) Dataset. Prepared by the authors of this paper TYolov5: A Temporal Yolov5 Detector Based on Quasi-Recurrent Neural Networks for Real-Time Handgun Detection in Video.
- Real-time gun detection in CCTV : An open problem.
- Detection-of-lumbar-weapon-pistol-by-deep-learning--With-YOLO-v5.
- https://www.kaggle.com/datasets/atulyakumar98/fire-and-gun-dataset/
AI For Mankind combined the following datasets into the above one consolidated train dataset and trained an experimental Yolov5 model for gun detection.
-
Hand, gun and phone dataset prepared by Mario Alberto Duran-Vega, Miguel Gonzalez-Mendoza, Leonardo Chang, Cuauhtemoc Suarez-Ramirez, the authors of TYolov5: A Temporal Yolov5 Detector Based on Quasi-Recurrent Neural Networks for Real-Time Handgun Detection in Video
-
Rifle, gun and fire dataset prepared by Atulya Kumar
Here are the experimental model's metrics against validation set. (Yolov5 large, 300 epochs, batch size 8, image size 640)
YOLOv5l_rifle summary: 267 layers, 46119048 parameters, 0 gradients, 107.7 GFLOPs
Class | Images | Instances | P | R | mAP50 | mAP50-95 |
---|---|---|---|---|---|---|
all | 500 | 645 | 0.834 | 0.741 | 0.77 | 0.496 |
gun | 500 | 195 | 0.898 | 0.774 | 0.837 | 0.658 |
fire | 500 | 151 | 0.798 | 0.722 | 0.744 | 0.349 |
rifle | 500 | 299 | 0.808 | 0.726 | 0.73 | 0.481 |
Our volunteers are working on a better model. Checkout our Colab notebook in the repo.