Dataset · MMInstruction
Illustrations and captions from arXiv research papers
MMInstruction/ArxivCap
Image-caption pairs from arXiv research publications with metadata for multimodal models.
What's inside
The dataset consists of images (charts, diagrams, illustrations) and corresponding captions extracted from English-language arXiv research articles. Each pair includes metadata on article category, license, and citation counts from Kaggle and Semantic Scholar. The collection volume ranges from 100 thousand to 1 million records.
What you can do with it
- Train VLM models to understand scientific charts and diagrams
- Generate descriptions for technical and scientific illustrations
- Multimodal search across research publications
- Tasks
- Image → text
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 15.3K
- Likes
- 58
How to load it
from datasets import load_dataset
ds = load_dataset("MMInstruction/ArxivCap")