SpatialRGPT: Grounded Spatial Reasoning in Vision Language Models

TL;DR

SpatialRGPT significantly improves vision language models' ability to understand and reason about 3D spatial arrangements by integrating regional representations and depth information, supported by a new benchmark for evaluation.

Contribution

Introduces SpatialRGPT with a data pipeline and plugin module for enhanced 3D spatial reasoning in VLMs, along with a new benchmark for evaluation.

Findings

Enhanced spatial reasoning performance in VLMs.

Strong generalization to complex spatial relations.

Effective as a dense reward annotator for robotics.

Abstract

Vision Language Models (VLMs) have demonstrated remarkable performance in 2D vision and language tasks. However, their ability to reason about spatial arrangements remains limited. In this work, we introduce Spatial Region GPT (SpatialRGPT) to enhance VLMs' spatial perception and reasoning capabilities. SpatialRGPT advances VLMs' spatial understanding through two key innovations: (1) a data curation pipeline that enables effective learning of regional representation from 3D scene graphs, and (2) a flexible plugin module for integrating depth information into the visual encoder of existing VLMs. During inference, when provided with user-specified region proposals, SpatialRGPT can accurately perceive their relative directions and distances. Additionally, we propose SpatialRGBT-Bench, a benchmark with ground-truth 3D annotations encompassing indoor, outdoor, and simulated environments, for evaluating 3D spatial cognition in VLMs. Our results demonstrate that SpatialRGPT significantly enhances performance in spatial reasoning tasks, both with and without local region prompts. The model also exhibits strong generalization capabilities, effectively reasoning about complex spatial relations and functioning as a region-aware dense reward annotator for robotic tasks. Code, dataset, and benchmark are released at https://www.anjiecheng.me/SpatialRGPT