Real world Anomaly Detection in Surveillance Videos

This repository is a re-implementation of “Real-world Anomaly Detection in Surveillance Videos” with pytorch. As a result of our re-implementation, we achieved a much higher AUC than the original implementation.

Datasets

Download following data link and unzip under your $DATA_ROOT_DIR.
/workspace/DATA/UCF-Crime/all_rgbs

  • Directory tree
   DATA/
       UCF-Crime/ 
           ../all_rgbs
               ../~.npy
           ../all_flows
               ../~.npy
       train_anomaly.txt
       train_normal.txt
       test_anomaly.txt
       test_normal.txt
       

train-test script

python main.py

Reslut

METHOD DATASET AUC
Original paper(C3D two stream) UCF-Crimes 75.41
RTFM (I3D RGB) UCF-Crimes 84.03
Ours Re-implementation (I3D two stream) UCF-Crimes 84.45

Visualization

sam

result-5

Acknowledgment

This code is heavily borrowed from Learning to Adapt to Unseen Abnormal Activities under Weak Supervision and AnomalyDetectionCVPR2018.

GitHub

https://github.com/seominseok0429/Real-world-Anomaly-Detection-in-Surveillance-Videos-pytorch

 

 

 

To finish reading, please visit source site