面向稀疏视频理解与推理
Towards Sparse Video Understanding and Reasoning
把「选哪些帧」变成多轮推理中的可学习决策：用对话历史作记忆、跨轮维护思维链、自纠错保证结构有效，可直接套到任意 VLM 的视频输入前端，替代均匀采样以在更少帧下提升 VQA 准确率。
原文依据 ReViSe interactively selects informative frames through multi-round reasoning.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论本文提出 ReViSe（Reasoning with Video Sparsity），一个将多轮推理与自适应帧选择相结合的视频问答（VQA）框架。现有的视觉语言模型（VLM）对视频帧进行均匀采样，这会引入冗余或无关信息。与之不同，ReViSe 通过多轮推理交互式地选择信息量大的帧。为此，ReViSe 包含三个模块：一个多轮对话模块，将帧选择历史保留为记忆；一个推理追踪器，在多个轮次之间维护思维链；以及一个自纠错机制，强制保证结构与行为上的有效性。ReViSe 可与专有和开源 VLM 无缝集成：它以「即插即用」的方式支持专有模型，并可为开源模型启用强化微调。在多个 VQA 基准上的实验表明，ReViSe 在减少所用帧数的同时提升准确率，从而增强了 VLM 的视频理解能力。
英文原文摘要 1035 字符
We present **ReViSe** (_**Re**asoning with **Vi**deo **S**parsity_), a framework that combines multi-round reasoning with adaptive frame selection for video question answering (VQA). Existing vision-language models (VLMs) uniformly sample video frames, which introduces redundancy or irrelevancy. In contrast, ReViSe*interactively selects informative frames through multi-round reasoning. To achieve this, ReViSe includes three modules: a multi-round conversation module that retains frame selection history as memory; a reasoning tracer that maintains a chain-of-thought across rounds; and a self-correction mechanism that enforces structural and behavioral validity. ReViSe integrates seamlessly with both proprietary and open-source VLMs. It supports proprietary models in a "plug-and-play" manner and enables reinforcement fine-tuning for open-source models. Experiments on multiple VQA benchmarks show that **ReViSe** improves the video understanding ability of VLMs by improving accuracy while reducing the number of frames used.