OmniCompass：全模态视频定位与理解整体基准
原标题：Paper page - OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
AI 摘要
研究团队提出 OmniCompass，一个面向全模态视频定位与理解的整体性基准，通过时序定位、空间定位和问答任务评估视觉-语言与音视频推理能力。该基准包含 OmniCompass-VL 和 OmniCompass-AV 两条评测轨道，共 820 个视频、2460 个实例，并配有 240 个视频、720 个实例的音视频子集，所有实例均经人工多轮质量校验。对 50 个多模态大模型的评测显示，开源模型仍落后于闭源模型，尤其在音视频理解方面，且精确的时序与空间定位对领先模型依然困难，预测在不同目标指定方式下存在不一致。
正文节选
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding Abstract OmniCompass is a holistic benchmark for omni-modal video grounding and understanding that evaluates vision-language and audio-visual reasoning through temporal grounding, spatial grounding, and question answering tasks. With the rapid development of multimodal large language models (MLLMs), evaluating their video understanding capabilities has become increasingly important. Existing video understanding be