Dataset · ReasonCore
Road camera 3D spatial reasoning benchmark
ReasonCore/open-spatial-reasoning
Multiple-choice visual question dataset for testing 3D spatial understanding in multimodal models.
What's inside
An English multimodal dataset of under 1000 examples combining car camera images and multiple-choice text questions. The images feature numbered bounding boxes around scene objects, and the questions test true 3D spatial reconstruction without relying on simple 2D heuristics like bounding box size or frame position.
What you can do with it
- Evaluating and comparing vision-language models in road scene 3D perception tasks.
- Testing autonomous driving systems for robustness against superficial visual cues.
- Tasks
- Multiple choice
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 875
- Likes
- 66
How to load it
from datasets import load_dataset
ds = load_dataset("ReasonCore/open-spatial-reasoning")