Dataset · allenai
Synthetic image-QA pairs
allenai/CoSyn-400K
A collection of generated images and question-answer pairs based on executable code from AllenAI.
What's inside
A multimodal dataset containing computer-generated images paired with questions and answers (volume ranges from 100K to 1M). The images were obtained by rendering code written by the Claude model, while the questions and answers were generated directly by GPT-4o mini from this code without viewing the image itself.
What you can do with it
- Training and fine-tuning multimodal models (VLMs) on Visual Question Answering tasks
- Training models to accurately read charts, diagrams, and synthetic visual scenes
- Evaluating neural network capabilities in answering questions about structured images
- Tasks
- Visual question answering
- Size
- 100K–1M rows
- Downloads this month
- 8.1K
- Likes
- 54
How to load it
from datasets import load_dataset
ds = load_dataset("allenai/CoSyn-400K")