#ModelAccuracyPaper
1n/3 worst (ours)52.10link
2n/2 worst50.40link
3n/4 worst46.80link
4Uniform mean46.30link
5Euclid's Gift: Enhancing Spatial Perception and Reasoning in Vision-Language Models via Geometric Surrogate Tasks41.80link
VSI-Bench vsi-bench Leaderboard