/I3D-PyTorchVideo

Primary LanguagePython

I3D-PyTorchVideo

Reference： https://github.com/tianyu0207/RTFM https://pytorchvideo.org/docs/tutorial_torchhub_inference

Download an vieo for utils/utils.py test.

wget -c https://dl.fbaipublicfiles.com/pytorchvideo/projects/archery.mp4

Download UCF-Crime dataset.

Website: https://www.crcv.ucf.edu/projects/real-world/

The downloaded dataset is shown below:

├── Anomaly_Train.txt
├── Anomaly-Videos-Part-1.zip
├── Anomaly-Videos-Part-2.zip
├── Anomaly-Videos-Part-3.zip
├── Anomaly-Videos-Part-4.zip
├── config.yaml
├── Normal_Videos_for_Event_Recognition.zip
├── ReadMe-Anomaly-Detection.txt
├── Temporal_Anomaly_Annotation_for_Testing_Videos.txt
├── Testing_Normal_Videos.zip
├── Training-Normal-Videos-Part-1.zip
├── Training-Normal-Videos-Part-2.zip
└── UCF_Crimes-Train-Test-Split.zip

After deleting Normal_Videos_for_Event_Recognition.zip, unzip all the files and organize them, the directory tree is as follows:

.
├── Anomaly_Detection_splits
├── Anomaly_Train.txt
├── Anomaly-Videos-Part-1
├── Anomaly-Videos-Part-2
├── Anomaly-Videos-Part-3
├── Anomaly-Videos-Part-4
├── Temporal_Anomaly_Annotation_for_Testing_Videos.txt
├── Testing_Normal_Videos_Anomaly
├── Training-Normal-Videos-Part-1
└── Training-Normal-Videos-Part-2

Instructions：

Anomaly-Videos-Part-{1..4} is the abnormal dataset, which contains 950 anomaly videos.
    Anomaly_Train.txt : Division of the training set of anomalous videos.
    Temporal_Anomaly_Annotation_for_Testing_Videos.txt : Division of abnormal video testing set.

Testing_Normal_Videos_Anomaly, Training-Normal-Videos-Part-1, and Training-Normal-Videos-Part-2 are the abnormal dataset, which contains 950 normal videos. The training and test datasets have been divided.

Feature extract

python main.py

After feature extraction，the directory tree is as follows:

├── Abuse
├── Arrest
├── Arson
├── Assault
├── Burglary
├── Explosion
├── Fighting
├── RoadAccidents
├── Robbery
├── Shooting
├── Shoplifting
├── Stealing
├── Testing_Normal_Videos_Anomaly
├── Training-Normal-Videos-Part-1
├── Training-Normal-Videos-Part-2
└── Vandalism

Merge Testing_Normal_Videos_Anomaly, Training-Normal-Videos-Part-1, and Training-Normal-Videos-Part-2:

cd feature_extract
mkdir Normal_Videos_event
cp Training-Normal-Videos-Part-1/* Training-Normal-Videos-Part-2/* Testing_Normal_Videos_Anomaly/* Normal_Videos_event
rm -rf Training-Normal-Videos-Part-1 Training-Normal-Videos-Part-2 Testing_Normal_Videos_Anomaly

Copy the label from repo: https://github.com/seominseok0429/Real-world-Anomaly-Detection-in-Surveillance-Videos-pytorch

Attention: The train_normal.txt has some repeating elements in it！

Feature extract Success!