CosmicAstro/UCF-Crime-X3D

UCF-Crime X3D Features

Dataset Description

This repository contains pre-extracted X3D spatio-temporal video features from the UCF-Crime dataset for research in video anomaly detection and weakly supervised video anomaly detection.

The features are intended to be used as input to downstream anomaly detection models rather than as raw video data.

Source Dataset

The features are extracted from videos belonging to the UCF-Crime dataset.

UCF-Crime is a large-scale surveillance video dataset containing normal and anomalous events from real-world surveillance footage.

The original dataset contains multiple anomaly categories, including:

• —Abuse
• —Arrest
• —Arson
• —Assault
• —Burglary
• —Explosion
• —Fighting
• —Road Accidents
• —Robbery
• —Shooting
• —Shoplifting
• —Stealing
• —Vandalism

as well as normal surveillance videos.

Feature Extraction

The videos were processed using an X3D spatio-temporal network to obtain compact video representations.

The resulting files contain pre-extracted spatio-temporal features and are intended for downstream video anomaly detection experiments.

Directory Structure

The repository is organized according to the corresponding UCF-Crime categories:

X3D_Videos/
├── Abuse/
├── Arrest/
├── Arson/
├── Assault/
├── Burglary/
├── Explosion/
├── Fighting/
├── RoadAccidents/
├── Robbery/
├── Shooting/
├── Shoplifting/
├── Stealing/
├── Testing_Normal_Videos_Anomaly/
├── Training_Normal_Videos_Anomaly/
└── Vandalism/