LLaVA-OVSI
收藏arXiv2025-09-30 收录
下载链接：
https://huggingface.co/datasets/lmms-lab/LLaVA-OneVision-Data
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含260万种多模态数据点，涵盖自然图像、数学与推理问题、文档、图表、屏幕截图以及通用OCR等，涉及超过90个数据来源。该数据集用于采样选择，并遵循一种健壮的多模态教学数据采集方法。其规模达到260万个实例，旨在支持单一图像针对多模态大型语言模型的微调任务。
This dataset contains 2.6 million multimodal data points, covering natural images, mathematical and reasoning problems, documents, charts, screenshots, general OCR content and other categories, with over 90 data sources. It is intended for sampling and selection, and adopts a robust multimodal instructional data collection methodology. With a total of 2.6 million instances, this dataset aims to support the fine-tuning of multimodal large language models using single-image inputs.
提供机构：
LLaVA Lab
