Dataset · shuolucs
Math spatial reasoning benchmark
shuolucs/MathSpatial
10,000 math problems with 26K+ geometric diagrams for evaluating MLLM spatial reasoning.
What's inside
MathSpatial is an open-source dataset benchmark in English and Chinese presented at ACM Multimedia 2026. It includes 10,000 math problems accompanied by over 26,000 geometric diagrams. The set is designed to test how well multimodal language models (MLLMs) understand space.
What you can do with it
- Evaluate and compare multimodal models (VQA) in geometry tasks
- Fine-tune MLLMs to solve math problems based on diagrams
- Investigate model hallucinations when analyzing spatial relationships
- Tasks
- Visual question answering
- Languages
- Chinese, English
- Size
- 10K–100K rows
- Downloads this month
- 20.3K
- Likes
- 3
How to load it
from datasets import load_dataset
ds = load_dataset("shuolucs/MathSpatial")