VideoLVR: Latent Visual Reasoning over Grounded Video Evidence.

arXiv:2610.37613v1 Announce Type: cross Abstract: Chain-of-thought reasoning has substantially advanced video understanding in multimodal large language models (MLLMs). However, language-based reasoning traces do not provide a reliable mechanism for models to