/viper

🐍 ViPeR is a framework built with TensorFlow for video and image classification. It was originally developed for the task of detecting violence in CCTV videos.

Primary LanguagePython

🐍 ViPeR: Video Pattern Recognition

ViPeR was designed to facilitate the training and evaluation of TensorFlow models for classifying videos/images of violence from security cameras, however, it can be used for a general purpose. The development was done during my bachelor's degree in Computer Science and paused during it. I intend to refactor all the code and include new features but the legacy version already does a lot if you want to use it.

Legacy/Stable version

All the source code is in the /legacy folder and it works fine (as far as I've tested it), you can find some usage examples in /legacy/demo.ipynb. For installation it is highly recommended to use Anaconda and run the following commands:

  1. conda create --name viper python=3.7
  2. conda activate viper
  3. conda install tensorflow-gpu==2.2.0
  4. conda install opencv pandas matplotlib
  5. pip install gdown

Next steps

  • Refactoring to be easier to use, similar to other frameworks like Detectron2
  • Integration with linting and testing tools
  • Integration with MLFlow
  • Integration with new models
  • Improve annotation tool
  • Add feature extraction module