论文概要
研究领域: CV 作者: Shijie Wang, Honglu Zhou, Ziyang Wang, Ran Xu, Caiming Xiong 发布时间: 2026-07-13 arXiv: 2607.11862
中文摘要
当前视频大语言模型（Video LLM）在问答方面表现出色，但主要作为黑箱运行，仅提供文本答案而无可验证的视觉 grounding。现有可解释性工作依赖文本理由或稀疏边界框，难以捕捉遮挡和非刚性形变等复杂视频动态。本文提出 Evidence-Backed Video QA（E-VQA），一种要求模型同时输出语义答案和精确时空证据的新任务：时间段和密集跟踪的物体分割掩码序列。为此引入ST-Evidence，首个针对判别性和生成性像素级grounding的人类验证基准。对SOTA模型的评估揭示问答准确性与真实视觉感知之间存在关键解耦，仅靠扩大规模无法弥合。本文开发了可扩展的自动化生成流程创建16万规模的ST-Evidence-Instruct数据集，在grounding Video LLM上微调后相比同规模UniPixel基线获得大幅提升（7B模型上t-mean+27.2，J&F+13.8），为可解释、基于证据的视频理解建立了稳健基线。
原文摘要
Current Video Large Language Models (Video LLMs) excel in question answering (QA) but largely operate as black boxes, providing textual answers without verifiable visual grounding. Existing explainability efforts rely on textual rationales or sparse bounding boxes, which struggle to capture complex video dynamics such as occlusions and non-rigid deformations. We propose Evidence-Backed Video Question Answering (E-VQA), a novel task requiring models to jointly output a semantic answer and precise spatio-temporal evidence: temporal segments and dense, tracked object segmentation masklets. To support this, we introduce ST-Evidence, the first human-verified benchmark for both discriminative and generative pixel-level grounding. Evaluations of state-of-the-art models reveal a critical decouplin…
— 自动采集于 2026-07-15
#论文 #arXiv #CV #小凯