VSI-Bench
A Vision-Language Spatial Intelligence Benchmark designed to evaluate a model's ability to understand and reason about spatial relationships under pure visual input conditions, without access to depth sensors or 3D annotations.
Timeline (last 90 days)
DK-VLM scored 74.33 on VSI-Bench and 58.28 on ReVSI-32, both the highest total scores.
VSI-Bench is a benchmark for video spatial intelligence, covering tasks such as object counting, relative orientation, path planning, and perspective transformation.
The authors suggest that developers recompute the latency budget on VSI-Bench, expecting to achieve over 90% reduction in execution time while the success rate is only a few percentage points lower than the original teacher.
SpatialOPSD uses validated programming agent execution trajectories and geometric descriptors to train a model that relies solely on RGB input, surpassing the SFT and GRPO baselines on the VSI-Bench benchmark.
All Facts (3)
DK-VLM scored 74.33 on VSI-Bench and 58.28 on ReVSI-32, both the highest total scores.
50%UnverifiedVSI-Bench is a benchmark for video spatial intelligence, covering tasks such as object counting, relative orientation, path planning, and perspective transformation.
50%UnverifiedSpatialOPSD uses validated programming agent execution trajectories and geometric descriptors to train a model that relies solely on RGB input, surpassing the SFT and GRPO baselines on the VSI-Bench benchmark.
50%