Bidirectional skip-frame prediction for video anomaly detection with intra-domain disparity-driven attention (BiSP)
submitted for the journal PR
python 3.8
torch 1.13
-
Data preparation
Ped1/Ped2: http://www.svcl.ucsd.edu/projects/anomaly/dataset.htm
Avenue: https://www.cse.cuhk.edu.hk/leojia/projects/detectabnormal/dataset.html
ShanghaiTech: https://svip-lab.github.io/dataset/campus_dataset.html -
Train
python Train.py # Ped2
python Train_ped1.py # Ped1
python Train_avenue.py # Avenue
python Train_shanghaitech.py # ShanghaiTech -
Evaluation
python Evaluate.py # Ped2
python Evaluate_ped1.py # Ped1
python Evaluate_avenue.py # Avenue
python Evaluate_shanghaitech.py # ShanghaiTech
You can download the pretrained weights of BiSP for the four datasets from Google.