Video Anomaly Detection with Spatio-Temporal Dissociation

Prerequisites

The code is built with following libraries: opencv-python Pillow scikit-learn scipy sklearn pytorch tqdm

Data Preparation

We have trained on Ped2, Avenue and ShanghaiTech

Training

python script.py 
    --pretrain_tag 1
    --ini_cluster_tag 1
    --finetune_tag 1
    --log_path log_dir
    --dataset_path datasets_dir/ShanghaiTechCampus  
    --dataset_name shanghai_tech 
    --eval 0

Evaluation

python script.py 
    --log_path log_dir
    --dataset_path datasets_dir/ShanghaiTechCampus  
    --dataset_name shanghai_tech 
    --eval 1

Pretrained Model Weights

We also provide the checkpoints of Ped2, Avenue and ShanghaiTech.