Collaborative Learning of Anomalies with Privacy (CLAP) for Unsupervised Video Anomaly Detection: A New Baseline

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)


Unsupervised (US) video anomaly detection (VAD) in surveillance applications is gaining more popularity recently due to its practical real-world applications. As surveillance videos are privacy sensitive and the availability of large-scale video data may enable better US-VAD systems, collaborative learning can be highly rewarding in this setting. However, due to the extremely challenging nature of the US-VAD task, where learning is carried out without any annotations, privacy-preserving collaborative learning of US-VAD systems has not been studied yet. In this paper, we propose a new baseline for anomaly detection capable of localizing anomalous events in complex surveillance videos in a fully unsupervised fashion without any labels on a privacy-preserving participant-based distributed training configuration. Additionally, we propose three new evaluation protocols to benchmark anomaly detection approaches on various scenarios of collaborations and data availability. Based on these protocols, we modify existing VAD datasets to extensively evaluate our approach as well as existing US SOTA methods on two large-scale datasets including UCF-Crime and XD-Violence.

PromptAlign Concept.

Architecture of CLAP, an unsupervised video anomaly detection model trained by multiple collaborating participants.

Example of Random Dataset Split

Random Dataset Split SVG

Example of Event Dataset Split

Event Dataset Split SVG

Example of Scene Dataset Split

Scene Dataset Split SVG

Dataset Distribution

PromptAlign Concept.

Distribution of UCF-Crime dataset videos based on the three training data organizations proposed in our paper to evaluate collaborative learning approaches for video Anomaly Detection.

CLAP Results

We evaluate CLAP on 3 different FL methods on the scene-based split
CLAP 73.99% 73.4% 73.7%

Comparison of CLAP with other SOTA Unsupervised and Weakly Supervised methods
Method UCF-Crime XD-Violence
Centralized GCL 71.04% 79.84% 73.62% 82.18%
C2FPL 80.65% 83.40% 80.09% 89.34%
CLAP 80.9% 85.50% 81.71% 90.04%
Local GCL 56.63% 65.32% 58.11% 59.91%
C2FPL 61.33% 65.85% 60.05% 63.4%
CLAP 63.93% 67.47% 62.37% 64.97%
Collaborative GCL 67.12% 76.82% 68.19% 75.21%
C2FPL 75.20% 77.60% 74.36% 76.98%
CLAP 78.02% 83.23% 77.65% 85.67%


      title={Collaborative Learning of Anomalies with Privacy (CLAP) for Unsupervised Video Anomaly Detection: A New Baseline},
      author={Al-lahham, Anas and Zaheer, Muhammad Zaigham and Tastan, Nurbek and Nandakumar, Karthik},
      journal={arXiv preprint arXiv:2404.00847},