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

TL;DR

A new baseline for anomaly detection capable of localizing anomalous events in complex surveil-lance videos in a fully unsupervised fashion without any labels on a privacy-preserving participant-based distributed training configuration is proposed.

Abstract

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. All proposed evaluation protocols, dataset splits, and codes are available here: https://github.com/AnasEmad11/CLAP

Figures (12)

• Figure 1: a) Conventional central training requires all training data to be on the server to carry out the training. This setting cannot ensure privacy, thus hindering collaborations between different entities holding large-scale surveillance data. b) Our proposed unsupervised video anomaly detection technique does not require the transfer of training data between the server and participants, thus ensuring complete privacy.
• Figure 2: Architecture of CLAP, an unsupervised video anomaly detection model trained by multiple collaborating participants.
• Figure 3: 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. (a) Random distribution of the videos is the baseline in which each participant has an almost identical number of videos and the classes are balanced. (b) Each participant holds videos containing certain types of anomalous events such as. shooting, robbery, etc. It is a relatively complex setting with the number of videos and class balance varying slightly between participants. (c) Each participant holds videos belonging to certain scenes such as shops, offices, etc. This is the most challenging setting where severe data and class imbalance are present across participants.
• Figure 4: AUC % performance of CLAP on UCF-Crime dataset using various proposed training splits under collaborative learning setting.
• Figure 5: Left: Comparison between GCL zaheer2022generative and CLAP with varying number of participants. Right: Number of participants with weak video-level supervision available.