real-time violence detection on video content.
This is a demo of a violence detection deep learning model. The model is based on Deepminds kinetics-i3d and was developed using transfer learning. The model was trained on a dataset of 35512 video clips (64 frame/clip) and produced the following results on the test data:
Accuracy | Precision | Recall |
---|---|---|
91.6% | 89.58% | 84.71% |
the code was tested on an nvidia GPU with cudnn drivers, it might not run properly on cpu .
clone the repository:
git clone https://github.com/null-void-Q/real-time-violence-detection.git
install requirments:
pip install -r requirments.txt
run the demo:
python main.py
after the local server is up head to localhost:5006 in your browser to use the web interface. (included test_video.mp4)