search

wanboyang / anomaly_detection_LAD2000

25 stars 2 forks source link

readme

Introduction

This repository is established for Anomaly detection in video sequences: A benchmark and computational model. In this repository, we provide a video anomaly detection database named LAD2000, which contains 2000 videos and 14 anomaly categories. The original paper can be found here.

If this work have some benefits to your research, please cite with the following BibTeX:

@article{wan2021anomaly,
  title={Anomaly detection in video sequences: A benchmark and computational model},
  author={Wan, Boyang and Jiang, Wenhui and Fang, Yuming and Luo, Zhiyuan and Ding, Guanqun},
  journal={IET Image Processing},
  year={2021},
  publisher={Wiley Online Library}
}

Requirements

conda env create -f environment.yaml

Data preparation

1.Downloading the original videos from (https://pan.baidu.com/s/1lQTqzqdA0dpi6Gl3ws9h5w pw:b1j8) or (link: https://stujxufeeducn-my.sharepoint.com/:f:/g/personal/2201810057_stu_jxufe_edu_cn/EqgRSGKhWJFNuKwak8CmC3QBC_Vp34KJ7vF48Fz7D_P3yA?e=K5YYOb pw：123456)

ps: The RGB and Flow frames of the original videos are provided in (https://stujxufeeducn-my.sharepoint.com/:f:/g/personal/2201810057_stu_jxufe_edu_cn/ErldMjSlsX1ElAlTgJbv8EsBiUCdwlz9O1xgqq07ktumfg?e=ksc1fv pw:123456)

2.Extract i3d visual features for LAD2000, you can clone this project (https://github.com/wanboyang/anomly_feature.pytorch) and just set --dataset to LAD2000.

or you can directly use the i3d features of our LAD2000

1.Download the i3d features(link: https://pan.baidu.com/s/1rzEfdY3PBND-5O1ScxTTUQ pw: jkjz ) or （link:https://stujxufeeducn-my.sharepoint.com/:f:/g/personal/2201810057_stu_jxufe_edu_cn/ElFomOTAEi1NsH_Oa63VYbQB0xrPMQIdNUaXLX3U-BHPkg?e=sViE5H pw:123456） and unzip the i3d.zip.

2.change the "dataset_path" to "you/path/i3d"

For LAD2000, Auenve, ped2, shanghaitech and UCF_Crime, we provide the full-supervised data splits and groundtruth in (link:https://stujxufeeducn-my.sharepoint.com/:u:/g/personal/2201810057_stu_jxufe_edu_cn/EQBo6YEqwLpMq0BwhIIu4KUBjZ5Cof2s96h_ebJQTCrcDA?e=fhtbs8 pw:123456 link:https://stujxufeeducn-my.sharepoint.com/:u:/g/personal/2201810057_stu_jxufe_edu_cn/EY22ebuTjM5LvmTIUlLjg2UBZCyskwMeNaCIu5zQjrNqHQ?e=PmYs8v pw:123456)

For Auenve, ped2, shanghaitech and UCF_Crime, we provide the i3d features in (link: https://pan.baidu.com/s/1fYAlFoTdcg8BgRdqoLQ2Bg pw: njy2) or (link:https://stujxufeeducn-my.sharepoint.com/:f:/g/personal/2201810057_stu_jxufe_edu_cn/EuQvbLCDoIxLgmcgpJRqcbIBzJSc7D6V-q151gLsWyFTrQ?e=JNbUEc pw:123456)

Class_index

class_dict = {'Drop': 3, 'Loitering': 9, 'Crash': 0, 'Violence': 13, 'FallIntoWater': 4, 'Fire': 7, 'Fighting': 6, 'Crowd': 1, 'Destroy': 2, 'Falling': 5, 'Trampled': 12, 'Thiefing': 11, 'Panic': 10, 'Hurt': 8}

Training

For LAD2000 database:

sh LAD2000T_i3d.sh

For ped2 database:

sh ped2_i3d.sh

For shanghaitech database:

sh ped2_i3d.sh

For Avenue database:

sh Avenue_i3d.sh

For UCF_Crime database:

sh UCF_i3d.sh

The models and testing results will be created on ./ckpt and ./results respectively

Acknowledgements

Thanks the contribution of W-TALC and awesome PyTorch team.

Contact

Please contact the first author of the associated paper - Boyang Wan （wanboyangjerry@163.com) for any further queries.