Learnable Locality-Sensitive Hashing for Video Anomaly Detection

This repository is the official implementation of Learnable Locality-Sensitive Hashing for Video Anomaly Detection.

Dependencies

• Hardware
  • CPU: 48 cores
  • RAM: 384 GB
  • Disk: 3 TB
  • GPU: NVIDIA GeForce 2080 Ti * 4
• Software
  • python 3.8.8 (in Anaconda)
  • cudatoolkit 11.1.1
  • pytorch 1.8.1
  • torchvision 0.9.1
  • numpy 1.19.2
  • scikit-learn 0.24.2
  • scipy 1.7.1
  • ffmpeg 4.3.2
  • opencv 4.5.3.56
  • slowfast 1.0 (https://github.com/facebookresearch/SlowFast)

Datasets

• Avenue: http://www.cse.cuhk.edu.hk/leojia/projects/detectabnormal/dataset.html
• ShanghaiTech: https://github.com/StevenLiuWen/sRNN_TSC_Anomaly_Detection
• Corridor: https://rodrigues-royston.github.io/Multi-timescale_Trajectory_Prediction

Usage

1. Extract features for the three datasets: 0-FeatureExtraction.
2. (Light-)LSH|LLSH: 1-LLSH.
3. KNN: 2-KNN.
4. K-means: 3-KMeans.

Please refer to the subdirectories for more details.

License

TODO

Related Repositories

• MoCo: https://github.com/facebookresearch/moco
• SlowFast: https://github.com/facebookresearch/SlowFast