wanboyang/ucf_2018_cvpr

A reproduce code for Real-world Anomaly Detection in Surveillance Videos

Introduction

This repository is a reproduce code for Real-world anomaly detection in surveillance videos (CVPR 2018). https://openaccess.thecvf.com/content_cvpr_2018/html/Sultani_Real-World_Anomaly_Detection_CVPR_2018_paper.html

Requirements

• Python 3
• CUDA
• numpy
• tqdm
• PyTorch (1.2)
• torchvision
  Recommend: the environment can be established by run

conda env create -f environment.yaml

Data preparation

1. Download the [c3d features][https://github.com/wanboyang/anomly_feature.pytorch].
2. Running the clip2segment.py get feature segments(32 segments for one video) and change the "dataset_path" to you/path/data

Training

python main.py --dataset_name UCF_Crime --feature_size 4096 --feature_modal rgb --feature_layer fc6

python main.py --dataset_name shanghaitech --feature_size 4096 --feature_modal rgb --feature_layer fc6

python main.py --dataset_name UCSDPed2 --feature_size 4096 --feature_modal rgb --feature_layer fc6 --k 32 --sample_size 4

python main.py --dataset_name ARD2000 --device 1 --feature_size 8192 --feature_modal rgb --feature_layer pool5 --feature_pretrain_model c3d

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

Acknowledgements

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