The research article presents a new framework for unsupervised video anomaly detection (VAD) called Coarse-to-Fine Pseudo-Labeling (C2FPL). Traditional VAD methods rely heavily on manually annotated anomaly examples, which is a laborious task. The proposed C2FPL framework aims to address this challenge by generating pseudo-labels to train an anomaly detector in a supervised manner. The framework employs hierarchical divisive clustering and statistical hypothesis testing to identify anomalous video segments from a set of completely unlabeled videos. The trained anomaly detector can then be applied to unseen test videos. The results show superior performance compared to existing methods, proving the effectiveness of the C2FPL framework.
Publication date: 26 Oct 2023
Project Page: https://github.com/AnasEmad11/C2FPL
Paper: https://arxiv.org/pdf/2310.17650
