fight_detection_yolov8

Fight/Violence Detection Dataset and Model Overview

This repository contains a dataset and YOLOv8 models (nano and small) trained to detect fights/violence and non-violence/no-fight in both videos and images. The models are optimized for surveillance and security applications where detecting physical confrontations is crucial.

Dataset Overview

• Dataset Classes: The dataset consists of two classes:

  1. Violence/Fight: Instances where physical violence is present.
  2. NoViolence/NoFight: Instances with no physical confrontations.

• Data Format:

  • Videos and Images are labeled accordingly for each class.
  • The dataset is designed for training deep learning models like YOLOv8 for violence detection.

Models

• YOLOv8 Models:
  • We have primarily trained YOLOv8-nano and YOLOv8-small models.
  • These models are lightweight and efficient, making them suitable for real-time detection tasks in resource-constrained environments.

Purpose

The models are trained to accurately detect violent events in various settings, including crowds, public spaces, and sports activities.

Key Features:

• Single Class Detection:
  • The attached code is specifically designed to detect one class at a time, with the focus being on the Violence/Fight class.
  • If the purpose is to detect only Violence/Fight, the models and code are pre-configured for this task.
  • Non-violence events are ignored during detection, allowing the model to concentrate solely on identifying violent actions.

Code and Usage Instructions

Pre-requisites:

• Python 3.8 or higher
• YOLOv8 (Ultralytics)
• PyTorch
• OpenCV

Running the Detection:

1. Clone the Repository:

   git clone (https://github.com/Musawer1214/Fight-Violence-detection-yolov8)
   cd <repository-directory>
   

2. Install Dependencies:

   pip install -r requirements.txt
   

3. Run Single Class Detection (Violence/Fight): The provided script detects only the Violence/Fight class in both videos and images.

   python detect.py --weights best.pt --source <input-video-or-image-path> --class 1 --save-txt
   

   Replace <input-video-or-image-path> with the path to the video or image you wish to analyze. The class ID for Violence/Fight is 1.

4. Model Weights:

   • The best.pt file contains the pre-trained YOLOv8-nano or YOLOv8-small model optimized for detecting violence/fights.

Notes

• Model Performance: The models are trained on a diverse set of images to generalize across different environments. However, additional fine-tuning may be required depending on your specific use case.
• Future Enhancements: We plan to extend the dataset and include more diverse scenarios to improve detection accuracy, including sports, public gatherings, and more.

---

License

This project is licensed under the MIT License.