FineVision

# Fine Vision


FineVision is a massive collection of datasets with **17.3M images**, **24.3M samples**, **88.9M turns**, and **9.5B answer tokens**, designed for training state-of-the-art open Vision-Language-Models.

More detail can be found in the blog post: https://huggingface.co/spaces/HuggingFaceM4/FineVision

### Load the data
```python
from datasets import load_dataset, get_dataset_config_names

# Get all subset names and load the first one
available_subsets = get_dataset_config_names('HuggingFaceM4/FineVision')
ds = load_dataset('HuggingFaceM4/FineVision', name=available_subsets[0], split='train', streaming=True)

# Inspect the first sample
ds[0]
```

### Structure

```bash
{
'images': [<PIL.PngImagePlugin.PngImageFile image mode=RGB size=387x194 at 0x7F8F0B308200>],
'texts': [{'user': 'Question: What is between the reticulum and the abomasum?\nChoices:\nA. Intestine\nB. Omasum\nC. Stomach\nD. Rumen\nAnswer with the letter.', 'assistant': 'Answer: B'},
{'user': 'Here is a diagram figure extracted from some Grade 1 - 6 science books.\nPlease first describe the content of this figure in detail, including how the knowledge visually displayed in the diagram.\nThen start with a section title "related knowledge:", briefly and concisely highlight the related domain knowledge and theories that underly this diagram. Note that you do not need to provide much detail. Simply cover the most important concepts.', 'assistant': "The figure is a simple diagram of the four compartments of a ruminant animal's stomach, which are the rumen, reticulum, omasum, and abomasum. The diagram shows the relative size and position of each compartment within the stomach, with arrows indicating the direction of food flow from one compartment to the next. \n\nRelated Knowledge:\n- Ruminant Digestion: Ruminants such as cows, sheep, and goats have a unique digestive system that allows them to break down fibrous plant material, like grass, that other animals cannot digest.\n- Four Stomach Compartments: The rumen is the largest compartment and serves as a fermentation vat where microbes break down fibrous material. The reticulum traps foreign objects and also helps in fermentation. The omasum absorbs water and nutrients, and the abomasum is the true stomach where digestion occurs similarly to monogastric animals.\n- Microbial Fermentation: The microbes in the rumen produce volatile fatty acids which are the primary energy source for ruminants. They also produce gases like methane, which are eructated (belched) out.\n- Ruminant Nutrition: Ruminants rely on a high-fiber diet and have to consume large quantities of forage to meet their nutritional needs."}],
'source': 'original',
'image_correspondence_ratings': [4, 3],
'image_correspondence_min': 3,
'visual_dependency_ratings': [4, 5],
'visual_dependency_min': 4,
'formatting_ratings': [4, 4],
'formatting_min': 4,
'relevance_ratings': [5, 5],
'relevance_min': 5
}
```

### Categories


### Licensing Information

Each of the publicly available sub-datasets present in FineVision are governed by specific licensing conditions. Therefore, when making use of them you must take into consideration each of the licenses governing each dataset. To the extent we have any rights in the prompts, these are licensed under CC-BY-4.0.

### Citation

If you find this dataset useful, please cite:
```
@misc{wiedmann2025finevisionopendataneed,
title={FineVision: Open Data Is All You Need},
author={Luis Wiedmann and Orr Zohar and Amir Mahla and Xiaohan Wang and Rui Li and Thibaud Frere and Leandro von Werra and Aritra Roy Gosthipaty and Andrés Marafioti},
year={2025},
eprint={2510.17269},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2510.17269},
}
```

# FineVision数据集


FineVision是一个超大规模数据集集合，涵盖**1730万张图像**、**2430万个样本**、**8890万轮交互**以及**95亿个答案Token**，专为训练顶尖开源视觉语言模型（Vision-Language Model）打造。

更多详情可参阅该博客文章：https://huggingface.co/spaces/HuggingFaceM4/FineVision

### 数据加载
python
from datasets import load_dataset, get_dataset_config_names

# 获取所有子数据集名称并加载第一个子集
available_subsets = get_dataset_config_names('HuggingFaceM4/FineVision')
ds = load_dataset('HuggingFaceM4/FineVision', name=available_subsets[0], split='train', streaming=True)

# 查看第一个样本
ds[0]

### 数据结构
bash
{
'图像': [<PIL.PngImagePlugin.PngImageFile 图像模式为RGB，尺寸387x194，存储地址为0x7F8F0B308200>],
'文本': [{'用户': '问题：网胃（reticulum）与皱胃（abomasum）之间的结构是什么？
选项：
A. 肠道（Intestine）
B. 瓣胃（Omasum）
C. 胃（Stomach）
D. 瘤胃（Rumen）
请以字母作答。', '助手': '答案：B'},
{'用户': '以下是取自1至6年级科学课本的示意图。
请首先详细描述该图的内容，包括该图以何种可视化方式呈现知识。
随后以章节标题“相关知识：”开头，简要提炼支撑该图示的相关领域知识与理论，无需展开过多细节，仅涵盖核心概念即可。', '助手': '该图为反刍动物胃的四个分区示意图，分别为瘤胃（rumen）、网胃（reticulum）、瓣胃（omasum）与皱胃（abomasum）。图示展示了各分区在胃内的相对大小与位置，并通过箭头标注了食物在各分区间的流动方向。

相关知识：
- 反刍动物消化：牛、绵羊、山羊等反刍动物拥有独特的消化系统，可分解其他动物无法消化的纤维性植物材料（如青草）。
- 胃的四个分区：瘤胃是最大的分区，作为发酵罐依靠微生物分解纤维物质；网胃可捕获异物并辅助发酵；瓣胃负责吸收水分与营养物质；皱胃是真正的胃，其消化机制与单胃动物相似。
- 微生物发酵：瘤胃中的微生物可产生挥发性脂肪酸，作为反刍动物的主要能量来源，同时还会产生甲烷等气体，通过嗳气排出体外。
- 反刍动物营养：反刍动物依赖高纤维饮食，需摄入大量饲草以满足营养需求。'}],
'来源': 'original',
'图像匹配评分': [4, 3],
'最低图像匹配评分': 3,
'视觉依赖评分': [4, 5],
'最低视觉依赖评分': 4,
'格式评分': [4, 4],
'最低格式评分': 4,
'相关性评分': [5, 5],
'最低相关性评分': 5
}

### 数据集类别

### 授权信息
FineVision中所有公开可用的子数据集均受特定授权条款约束。因此，在使用这些子数据集时，您必须遵守各自对应的授权协议。就本项目中提示词的相关著作权而言，其授权协议为CC-BY-4.0。

### 引用方式
如果您认为本数据集对您的研究有所帮助，请引用如下文献：

@misc{wiedmann2025finevisionopendataneed,
title={FineVision: Open Data Is All You Need},
author={Luis Wiedmann and Orr Zohar and Amir Mahla and Xiaohan Wang and Rui Li and Thibaud Frere and Leandro von Werra and Aritra Roy Gosthipaty and Andrés Marafioti},
year={2025},
eprint={2510.17269},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2510.17269},
}