jackyccl / MGFN.

This is the official repo of paper accepted in AAAI 2023 Oral.

MGFN:Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detection

This the official repo of paper accepted in AAAI 2023 Oral

Pretrained models can be downloaded in here:Onedrive

Prepare the environment:

    $pip install -r requirement.txt

gt generate reference github

Test: Download the pretrained model and run $test.py

Dataset Prepare: UCF-crime ten-crop I3D. Rename the data path in ucf-i3d.list and ucf-i3d-test.list based on your data path.

Train: Modify the option.py and run main.py

32 segmented UCF dataset: UCF-crime_ten-crop_I3D_32_seg. Rename the data path in ucf-i3d.list and ucf-i3d-test.list based on your data path.

Train: Use preprocessed flag if using the 32 seg data

Citation

If you find this repo useful for your research, please consider citing our paper:

  @misc{chen2022mgfn,
  title={MGFN: Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detection}, 
  author={Yingxian Chen and Zhengzhe Liu and Baoheng Zhang and Wilton Fok and Xiaojuan Qi and Yik-Chung Wu},
  year={2022},
  eprint={2211.15098},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
  }