VideoLVR: Latent Visual Reasoning over Grounded Video Evidence
- Field: Computer Vision / Video Reasoning
- Authors: Shilong Li, Xingyuan Bu, Hao Shi, Yun Xing, Weixin Feng, Qiangpeng Yang, Shilei Wen
- arXiv: 2610.37613
- Referring to evidence using textual coordinates, which specify symbolic locations rather than the visual content itself.
- Reinserting cropped regions or frames into the multimodal context, preserving visual details at the cost of substantial context expansion.
- A visual token prediction objective formulates visual token reconstruction as prediction over a vocabulary derived from the input video, unifying visual evidence reconstruction and text generation within the same token prediction framework.
- This shared framework enables MLLMs to reason over interleaved text tokens and latent visual states while integrating evidence from different moments.
- The method is further extended to reinforcement learning: sampled visual tokens are treated as actions alongside text tokens, jointly optimizing the policy with final-answer rewards.
Overview
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 revisit relevant visual evidence, leading to visual forgetting and impairing their ability to reason over video content.
Existing approaches maintain visual grounding during reasoning in two main ways:
The VideoLVR Approach
To address these limitations, VideoLVR constructs latent visual evidence from local regions at relevant video moments, providing visual support for intermediate reasoning steps.
Results
By efficiently revisiting task-relevant visual information during reasoning, VideoLVR improves video understanding without substantially expanding the multimodal context. Experiments on multiple video benchmarks demonstrate its effectiveness, especially on tasks requiring fine-grained visual evidence.
---
*Auto-collected on 2026-10-29.*