Search papers, labs, and topics across Lattice.
This paper introduces Video-DeepResearch (Video-DR), a multimodal agent designed to process continuous video streams, addressing critical issues of modality bias and parametric knowledge leakage in existing models. By implementing a decoupled perception-exploration pipeline and a two-stage training approach that includes supervised fine-tuning and Group Relative Policy Optimization (GRPO), the authors enable the agent to achieve superior performance in visual question answering. Empirical evaluations show that Video-DR outperforms leading models, achieving a state-of-the-art accuracy of 64.0% on a newly curated benchmark of complex multi-hop VQA tasks.
Video-DeepResearch shatters the imitation-learning ceiling, achieving a 64% accuracy in complex video-based question answering, far surpassing existing models.
We introduce Video-DeepResearch (Video-DR), extending multimodal agents from static images to continuous video streams, a setting that demands dense spatiotemporal grounding coupled with open-web exploration. Preliminary evaluations reveal two critical bottlenecks in current models: (1) modality bias, where agents bypass visual tools in favor of textual search, and (2) parametric knowledge leakage, where models rely on internal memory rather than genuine tool-augmented execution. To address these challenges, we propose Video-DR, featuring a decoupled perception-exploration pipeline with stage-wise tool unlocking that compels exhaustive cross-frame visual grounding prior to web retrieval. Our framework adopts a two-stage training recipe: supervised fine-tuning followed by Group Relative Policy Optimization (GRPO), enabling autonomous exploration that breaks the imitation-learning ceiling. Furthermore, we curate Video-DR-Bench, a human-AI collaborative benchmark comprising 200 complex, multi-hop VQA instances. Empirical results demonstrate that our Video-DeepResearch-35B-A3B establishes a new state-of-the-art of 64.0% average accuracy, surpassing proprietary Claude-4.5-Sonnet (59.0%) by 5.0 points and significantly outperforming GPT-5 (52.5%) and Gemini 2.5 Pro (57.5%). The 30B-A3B variant achieves 59.3%, competitive with Claude-4.5-Sonnet and demonstrating the effectiveness of our training paradigm even at compact scale. Code: https://github.com/Osilly/Vision-DeepResearch.
