PRIDE (Pseudo-label Refinement with Intensive Knowledge transfer and DisEntangled attention)

This repository contains the official Pytorch implementation of our paper: Intensive Knowledge Transfer for Weakly-Supervised Video Anomaly Detection.

• Authors: Jash Dalvi, Gunjan Dhanuka, Ali Dabouei, Min Xu
• Achieves state-of-the-art results on UCF-Crime and ShanghaiTech datasets.

Setting Up

• Clone the repository and navigate into the repo.

  git clone https://github.com/GunjanDhanuka/PRIDE
  cd PRIDE/
  

• Please download the required files from the drive link: Drive Link, extract them and place in the data folder.

• The final directory structure should look like -

  |--PRIDE/
      |-- config/
      |-- data/
          |-- I3D
              |-- all_rgbs/
              |-- all_flows/
          |-- S3D/
          |-- VideoSwin/
      |-- DataLoaders/
      |-- files/
      |-- Losses/
      |-- Models/
      |-- Utils/
      ...
  

• Change the file locations in config/config.yaml file to the absolute path of the folders on your device.

• Create a new virtual environment using your preferred method and install the required dependencies from the requirements.txt file. Note that we recommend installing PyTorch using the commands given on the PyTorch website, and then installing the rest of the packages using pip.