研究人员开发了一种名为语义证据奖励（SER）的新方法，以提高视频多模态大语言模型（Video MLLMs）的时空推理能力。现有模型在细粒度推理方面常常遇到困难，有时会使用不相关的帧或对象来回答问题。SER通过将证据定位重构为验证任务来解决这个问题，使用一个裁判VLM来评估生成证据的相关性和定位质量，从而减少对密集标注的需求。这种方法通过在V-STAR基准上提高3.0个点来增强答案准确性和证据定位，如所证明的。 AI
影响 增强了Video MLLM的准确性和证据定位，可能减少视频问答任务对大量标注的依赖。
排序理由 该集群包含多篇arXiv论文，详细介绍了一种用于改进Video MLLMs的新研究方法。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 7 个来源。 我们如何撰写摘要 →
arXiv:2508.07683v2 Announce Type: replace-cross Abstract: Video Temporal Grounding (VTG) aims to localize specific video segments corresponding to natural language queries. While recent Large Vision-Language Models (LVLMs) employ Reinforcement Learning to generate Chains-of-Thoug…
Video MLLMs often struggle with fine-grained spatio-temporal reasoning, sometimes generating correct answers based on irrelevant frames or objects. Although outputting spatio-temporal evidence during reasoning is a promising direction, existing RL frameworks typically rely on geo…
arXiv:2607.00672v1 Announce Type: new Abstract: Zero-shot video temporal grounding (VTG) localizes events in untrimmed videos from natural language queries without task-specific training. Existing methods rely on frame-query feature matching, which suffices for simple events but …
Zero-shot video temporal grounding (VTG) localizes events in untrimmed videos from natural language queries without task-specific training. Existing methods rely on frame-query feature matching, which suffices for simple events but struggles with complex multi-stage queries that …
arXiv:2603.05663v3 Announce Type: replace Abstract: Video Temporal Grounding (VTG) localizes the temporal boundaries of query-relevant moments in long, untrimmed videos, making video-language-model prohibitively expensive. While recent training-free token pruning has shown succes…
arXiv:2606.24726v1 Announce Type: new Abstract: Video MLLMs often struggle with fine-grained spatio-temporal reasoning, sometimes generating correct answers based on irrelevant frames or objects. Although outputting spatio-temporal evidence during reasoning is a promising directi…
Video MLLMs often struggle with fine-grained spatio-temporal reasoning, sometimes generating correct answers based on irrelevant frames or objects. Although outputting spatio-temporal evidence during reasoning is a promising direction, existing RL frameworks typically rely on geo…