PiercingEye: Dual-Space Video Violence Detection with Hyperbolic Vision-Language Guidance

2025-04-26

AI Summary

Existing weakly supervised video violence detection methods rely on Euclidean representations, which struggle to distinguish fine-grained violent events that are visually similar but semantically distinct. To address this, we propose the first Euclidean–hyperbolic dual-space learning framework. Our method introduces a layer-sensitive hyperbolic aggregation strategy and a cross-space attention mechanism to model hierarchical semantics; leverages large-language-model-generated logic-guided fuzzy event descriptions to construct a dynamically similarity-weighted hyperbolic vision–language contrastive loss; and imposes a hyperspherical Dirichlet energy constraint to enhance the discriminability of hyperbolic embeddings. Evaluated on XD-Violence and UCF-Crime, our approach achieves state-of-the-art performance, particularly demonstrating significant improvements in fine-grained recognition accuracy on a newly constructed subset of ambiguous violent events.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Machine Learning: Multi-instance/Multi-view Learning
• Natural Language Processing: Safety and Robustness

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing weakly supervised video violence detection (VVD) methods primarily rely on Euclidean representation learning, which often struggles to distinguish visually similar yet semantically distinct events due to limited hierarchical modeling and insufficient ambiguous training samples. To address this challenge, we propose PiercingEye, a novel dual-space learning framework that synergizes Euclidean and hyperbolic geometries to enhance discriminative feature representation. Specifically, PiercingEye introduces a layer-sensitive hyperbolic aggregation strategy with hyperbolic Dirichlet energy constraints to progressively model event hierarchies, and a cross-space attention mechanism to facilitate complementary feature interactions between Euclidean and hyperbolic spaces. Furthermore, to mitigate the scarcity of ambiguous samples, we leverage large language models to generate logic-guided ambiguous event descriptions, enabling explicit supervision through a hyperbolic vision-language contrastive loss that prioritizes high-confusion samples via dynamic similarity-aware weighting. Extensive experiments on XD-Violence and UCF-Crime benchmarks demonstrate that PiercingEye achieves state-of-the-art performance, with particularly strong results on a newly curated ambiguous event subset, validating its superior capability in fine-grained violence detection.

Problem

• Detect video violence with dual-space learning framework
• Enhance feature representation using Euclidean and hyperbolic geometries
• Generate ambiguous event descriptions via large language models

Innovation

• Dual-space learning with Euclidean and hyperbolic geometries
• Layer-sensitive hyperbolic aggregation with Dirichlet constraints
• Hyperbolic vision-language contrastive loss for ambiguous samples