Search papers, labs, and topics across Lattice.
This paper introduces Video-OPSD, an innovative framework that enhances on-policy self-distillation in Video Large Language Models by leveraging privileged visual evidence from annotated frames. By employing an Evidence-Grounded Self-Teacher that focuses on these frames while allowing the student to process the entire video, the method provides more targeted supervision. Experimental results demonstrate that Video-OPSD outperforms standard OPSD and achieves comparable performance to GRPO with significantly reduced training time, marking a substantial advancement in video reasoning capabilities.
Video-OPSD reveals that focusing on privileged visual evidence can drastically improve the efficiency and effectiveness of self-distillation in Video-LLMs.
On-policy self-distillation (OPSD) has recently emerged as an effective post-training paradigm that improves policy optimization through dense token-level supervision from a privileged self-teacher. Despite its promise, OPSD remains largely underexplored for Video Large Language Models (Video-LLMs). Existing methods typically construct privileged teachers by augmenting their context with additional information while keeping the primary input unchanged for both teacher and student. Video reasoning, however, offers a distinct source of privileged supervision within the primary input itself: long videos contain substantial temporal redundancy, and only a small subset of frames provides the evidence necessary to answer a question. Building on this observation, we present $\textbf{Video-OPSD}$, an OPSD framework that exploits privileged visual evidence for both self-teacher construction and knowledge transfer. First, our Evidence-Grounded Self-Teacher conditions the teacher exclusively on annotated evidence frames while the student continues to reason over the complete video. This focused visual input enables the teacher to provide more informative supervision. Second, our Evidence-Guided Token Optimization adaptively weights token-level distillation according to each reasoning token's reliance on privileged visual evidence, thereby emphasizing perceptually grounded reasoning. Experiments across video understanding and reasoning benchmarks show that $\textbf{Video-OPSD}$ consistently improves upon Standard OPSD across multiple backbones and achieves performance comparable to GRPO while requiring substantially less training time, establishing an effective and efficient post-training approach for Video-LLMs.
