Spatial Understanding from Videos: Structured Prompts Meet Simulation Data

TL;DR

This paper introduces a new framework that enhances 3D spatial reasoning in pre-trained vision-language models by combining structured prompts and a scalable simulation-based question-answering dataset, improving their understanding of complex scenes.

Contribution

It proposes a novel prompting strategy and a large-scale dataset to improve 3D spatial reasoning in existing vision-language models without altering their architecture.

Findings

Enhanced spatial reasoning performance across multiple benchmarks.

Structured prompts improve interpretability and reasoning accuracy.

Fine-tuning with the proposed dataset yields significant gains.

Abstract

Visual-spatial understanding, the ability to infer object relationships and layouts from visual input, is fundamental to downstream tasks such as robotic navigation and embodied interaction. However, existing methods face spatial uncertainty and data scarcity, limiting the 3D spatial reasoning capability of pre-trained vision-language models (VLMs). To address these challenges, we present a unified framework for enhancing 3D spatial reasoning in pre-trained VLMs without modifying their architecture. This framework combines SpatialMind, a structured prompting strategy that decomposes complex scenes and questions into interpretable reasoning steps, with ScanForgeQA, a scalable question-answering dataset built from diverse 3D simulation scenes through an automated construction process designed for fine-tuning. Extensive experiments across multiple benchmarks demonstrate the individual and combined effectiveness of our prompting and fine-tuning strategies, and yield insights that may inspire future research on visual-spatial understanding.