R1-Onevision 多模态推理数据集

R1-Onevision 数据集是由浙江大学于 2025 年发布的，旨在赋予模型高级的多模态推理能力，相关 Blog 为「R1-Onevision：Open-Source Multimodal Large Language Model with Reasoning Ability」。它通过在自然场景、科学、数学问题、基于 OCR 的内容和复杂图表等多个领域的丰富、上下文感知的推理任务，弥合视觉和文本理解之间的差距。

The R1-Onevision dataset was released by Zhejiang University in 2025, with the objective of equipping models with advanced multimodal reasoning capabilities. Its associated blog post is titled 'R1-Onevision: Open-Source Multimodal Large Language Model with Reasoning Ability'. This dataset bridges the gap between visual and textual understanding through rich, context-aware reasoning tasks across multiple domains including natural scenes, scientific content, mathematical problems, OCR-based content, and complex charts.

搜集汇总

背景与挑战

背景概述

R1-Onevision多模态推理数据集由浙江大学于2025年发布，旨在增强模型的高级多模态推理能力。该数据集通过涵盖自然场景、科学问题、数学内容、OCR文本和复杂图表等多样领域的推理任务，以弥合视觉与文本理解之间的鸿沟。