Weakly supervised video anomaly detection based on hyperbolic space.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Weakly supervised video anomaly detection methods are proliferating.
• Current methods are limited by Euclidean space, hindering complex pattern modeling and long-term context.
• This leads to misjudgments of anomalous events due to insufficient video representation.

Purpose of the Study:

• To explore the potential of hyperbolic space for enhancing weakly supervised video anomaly detection.
• To develop a novel method for transforming video features into hyperbolic space.
• To improve the network's ability to distinguish between normal and anomalous videos.

Main Methods:

• Transforming video features from Euclidean space to hyperbolic space.
• Utilizing hyperbolic space to model implicit relationships in video data.
• Developing a weakly supervised learning framework for anomaly detection.

Main Results:

• The proposed method achieves state-of-the-art performance on the XD-Violence dataset using only RGB information.
• The approach demonstrates superior ability in distinguishing normal and anomalous video events.
• The method requires the lowest number of parameters compared to existing approaches.

Conclusions:

• Transforming video features into hyperbolic space is a promising direction for weakly supervised video anomaly detection.
• The proposed hyperbolic approach effectively captures complex patterns and long-term context.
• This method offers an efficient and high-performing solution for video anomaly detection.