Geoaux is a geometry and spatial reasoning benchmark for vision-language models.
metadata.jsonl: metadata file for the Hugging Face dataset viewer
geoaux.json: full benchmark annotations in JSON list format
image/: problem images
index: sample identifier
knowledge: high-level knowledge category
answer: ground-truth answer
question: problem text
aux_think: auxiliary reasoning annotation
Visual Operation:… See the full description on the dataset page:
https://huggingface.co/datasets/cyborvirtue/Geoaux.