SpatialQuery: Benchmarking Geometry-Grounded Multi-Instance Spatial Reasoning in Vision-Language Models

2026-08-03

AI Summary

This work addresses the poor performance of existing vision-language models in metric spatial reasoning tasks involving multiple instances of the same category, where accurately identifying the nearest neighbor and its ground-truth distance remains challenging. The authors propose a training-free framework that enhances spatial reasoning without modifying model architecture or fine-tuning. It first applies Scene Cubifying to transform objects in an image into canonical bird’s-eye-view blocks of uniform size, each encoded with category information. This representation is then coupled with an Uncertainty-Aware Chain-of-Thought prompting mechanism to improve reasoning fidelity. Evaluated on the SPATIALQUERY-1M benchmark, the method achieves state-of-the-art results with Qwen3-VL-8B, yielding a Floor-MAE of 0.259 meters, an Uncertainty-Aware Accuracy@0.3m of 90.5%, and a nearest-neighbor identification accuracy of 84.18%, outperforming both specialized models and leading closed-source systems.

Abstract

Vision-language models (VLMs) achieve strong semantic understanding but remain unreliable in metric spatial reasoning, particularly when queries require comparing multiple instances of the same object category. We study this problem through the Closest-Instance Distance Query (CIDQ), where a model must identify the nearest visible candidate to a unique reference object and estimate their gravity-aligned floor-plane distance. We introduce SPATIALQUERY, a training- free framework for CIDQ reasoning from a single RGB image, together with SPATIALQUERY-1M, a benchmark containing over one million RGB-only question-answer pairs from 200 indoor scenes. SPATIALQUERY recovers instance-level metric geometry and transforms it into a canonical Bird's-Eye View through Scene Cubifying, which represents objects as uniformly sized, category-coded blocks to emphasize their relative floor- plane locations. We further propose Uncertainty-Aware Chain-of-Thought (UA-CoT) prompting, which incorporates geometry- derived per-instance uncertainty into the VLM reasoning process. Without task-specific fine-tuning or architectural modification, SPATIALQUERY with Qwen3-VL-8B achieves a Floor-MAE of 0.259 m, an Unc-Acc@0.3 m of 90.5%, and a proximity-decision accuracy of 84.18%, outperforming fine-tuned spatial specialists, general-purpose VLMs, and closed-source frontier models. Code, benchmark resources, and an interactive demo are available at https://namhai1810.github.io/SpatialQuery/.

Problem

• spatial reasoning
• vision-language models
• metric geometry
• multi-instance comparison
• distance estimation

Innovation

• Scene Cubifying
• Uncertainty-Aware Chain-of-Thought
• Metric Spatial Reasoning
• Bird's-Eye View
• Training-Free Framework