Dataset · 1fanj
Visual chart analysis benchmark
1fanj/Chartographer
English chart-question pairs to evaluate whether VLMs actually analyze images instead of guessing.
What's inside
The dataset contains charts and corresponding questions (1,000 to 10,000 examples) in English. Each sample family includes an original chart, its reconstructed version, and ten modified variants while preserving chart and question identifiers. This approach tests whether models rely on visual reasoning when answering.
What you can do with it
- Evaluating multimodal models (VLM) for robustness to chart data changes
- Testing visual question answering (VQA) systems without relying on memorized templates
- Tasks
- Visual question answering, Question answering
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 23.7K
- Likes
- 3
How to load it
from datasets import load_dataset
ds = load_dataset("1fanj/Chartographer")