FS-UCF-Crime: A Temporally Annotated Extension of UCF-Crime for Online Fully Supervised Video Anomaly Detection

FS-UCF-Crime is a fully supervised extension of the UCF-Crime dataset designed to support online video anomaly detection in surveillance scenarios. The dataset enriches anomalous UCF-Crime videos with interval-level temporal annotations. Each anomalous event is represented by its anomaly class, start time, and end time. Training videos are annotated from scratch, while the temporal annotations of the original UCF-Crime test split are reviewed, corrected where necessary, and completed with interval-level anomaly labels. FS-UCF-Crime preserves the original UCF-Crime train/test organization and introduces a validation partition used for model selection and threshold calibration. The annotations support fully supervised training on partial temporal chunks, temporal anomaly localization, and evaluation at chunk, event, and video levels. The complete annotation package will be released after acceptance of the associated paper, “Online Fully Supervised Video Anomaly Detection.” The release will include temporal annotations, interval-level class labels, validation split information, annotation documentation, and scripts for reproducing the training and evaluation protocol. The original UCF-Crime videos are not redistributed through this record and remain subject to the terms of use of their original sources.