/MIST_VAD

Official codes for CVPR2021 paper "MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection"

Primary LanguagePython

MIST_VAD

PWC PWC

Official codes for CVPR2021 paper "MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection"

Project Page

Paper

Structure of MIST

Updates

[May 28th] Testing / Training codes have been released. The codes are cleaned out from the original ones without full verification. There maybe any unexpected bugs. I will improve it later if I have time.

Requirements

  • python>=3.6
  • apex
  • pytorch=1.5.0+cu101
  • torchvision=0.6.0+cu101
  • tensorboardX
  • h5py
  • opencv
  • scikit-learn
  • yacs

Testing

Pretrained models have been uploaded on OneDrive.

The h5py file for ShanghaiTech and its corresponing annotations are uploaded on [BaiduYun] with multiple sub-files, you can open/unzip it with WinRAR

BaiduYun link, code:kym5

To test the pretrained checkpoints, you are recommended to read Testing_Guidelines.md for more details.

Training

We have released the training codes for ShanghaiTech and UCF-Crimes. For convenience to repeat our experiments, we presents the pseudo labels files in data/ dir. The details of training are listed in Training_Guidelines.md.

Reference

If you feel the codes help, please cite our paper.

Recommended Citation Form:

Jia-Chang Feng, Fa-Ting Hong and Wei-Shi Zheng. “MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection, Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition. 2021.