Unleashing Multimodal Large Language Models for Training-free HOI Detection in the Wild

arXiv cs.AI · 18 hours ago

Researchers developed AgentHOI, a training-free framework that uses multimodal large language models to detect human-object interactions without requiring labeled training data. The method uses context-aware multi-round reasoning and instance-specific descriptions to identify interactions in complex real-world scenes, achieving performance superior to supervised methods. This approach enables better generalization to open-world scenarios compared to traditional supervised detection systems that are limited to predefined interaction categories.