VicKAM: Visual Conceptual Knowledge Guided Action Map for Weakly Supervised Group Activity Recognition

• DOI: 10.48550/arxiv.2502.09967
• Publication Date: 2025-02-14
• Authors: Zhuming Wang; Yihao Zheng; Jiarui Li; Yu‐Te Wu; Yan Huang; Zun Li; Lifang Wu; Ao Wang
• Keywords: Action Recognition

Existing weakly supervised group activity recognition methods rely on object detectors or attention mechanisms to capture key areas automatically. However, they overlook the semantic information associated with captured areas, which may adversely affect performance. In this paper, we propose a novel framework named Visual Conceptual Knowledge Guided Action Map (VicKAM) effectively captures locations of individual actions and integrates them action semantics for recognition.It generates prototypes from training set as visual conceptual knowledge bridge representations. by knowledge, VicKAM produces maps that indicate likelihood each occurring at various locations, based image correlation theorem. It further augments using related statistical information, representing distribution under different activities, establish connections between specific activities. The augmented map is incorporated representations recognition.Extensive experiments two public benchmarks, Volleyball NBA datasets, demonstrate effectiveness our proposed method, even in cases limited data. code will be released later.