SOURCE-LINKED INTELLIGENCE

Where to Look Matters: On-Policy Self-Distillation for Long-Video Understanding

arXiv · AI, language, vision and robotics · article · Aug 26, 2026 · UTC

Vision-language models (VLMs) have made substantial progress in long-video understanding, with standard backbone models typically answering questions from frames sampled across the full video. However, as videos become longer, the full-video context inevitably contains more question-irrelevant temporal content, which can distract the model from the evidence needed to answer a specific question. We empirically find that focusing the visual input on short annotated clue intervals containing question-relevant evidence consistently improves prediction accuracy across model scales compared with usi

Read original source ↗ Open in workspace

recordType

paper

region

Global