Tracking the Truth: Object-Centric Spatio-Temporal Monitoring for Video Large Language Models

Tri Cao, Khoi Le, Thong Nguyen +7

While multimodal large language models (MLLMs) have advanced video understanding, they remain highly prone to hallucinations in dynamic scenes. We argue this stems from a failure i…