SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models
SpaceCast-Bench tests predictive spatial reasoning; the best of 21 models scores 58.0% versus 87.2% for humans.
SpaceCast-Bench evaluates predictive spatial reasoning with 3,862 questions drawn from 182 real-world scenes across 16 task types and three levels: static perception, local prediction, and global prediction. Among 21 models, the strongest reaches 58.0% against 87.2% human performance, while spatially specialized models remain near chance. Bridge views help integrate distributed observations, and explicit 3D evidence helps more reliably than generated outcome images or videos. Fine-tuning on programmatically generated data raises Qwen3-VL-4B from 34.0% to 65.7%, with gains on six out-of-domain benchmarks.
- The benchmark has 3,862 questions from 182 real-world scenes.
- The best of 21 models scores 58.0% versus 87.2% for humans.
- Explicit 3D evidence helps more than generated outcome images or videos.
- Fine-tuning lifts Qwen3-VL-4B from 34.0% to 65.7%.
Full article165 words · extracted from arxiv.org · click to collapse
Existing spatial reasoning benchmarks mainly test spatial perception: reading off relations already visible in the input. Yet real-world spatial intelligence demands predictive spatial reasoning: constructing a scene from observations, anticipating how an intervention changes it, and reasoning about the unseen outcome. We introduce SpaceCast-Bench, the first benchmark to directly and diagnostically evaluate this capability. Built around an observe-transform-infer framework, its 3,862 questions from 182 real-world scenes span 16 task types at three levels: static perception, local prediction, and global prediction, progressively requiring scene understanding, spatial state updating, and relational inference over unobserved outcomes. Evaluating 21 models exposes a stark gap: the strongest model reaches only 58.0% against 87.2% human performance, while spatially specialized models remain near random chance. Controlled analyses further reveal that bridge views are critical for integrating distributed observations, and that explicit 3D evidence benefits models more reliably than generated outcome images or videos. Fine-tuning on our programmatically generated data lifts Qwen3-VL-4B from 34.0% to 65.7% with macro-average gains across six out-of-domain benchmarks.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2610.12402