GroundSet: A Cadastral-Grounded Dataset for Spatial Understanding with Vector Data

TL;DR

GroundSet introduces a large-scale cadastral-based dataset for improving fine-grained spatial understanding in Earth Observation, enabling models to better interpret high-resolution aerial imagery for critical applications.

Contribution

The paper presents a new extensive dataset grounded in cadastral vector data, along with a benchmark demonstrating how supervision improves spatial reasoning in models.

Findings

Current models struggle with zero-shot spatial reasoning.

Supervised training significantly enhances model performance.

A standard LLaVA architecture serves as a robust baseline.

Abstract

Precise spatial understanding in Earth Observation is essential for translating raw aerial imagery into actionable insights for critical applications like urban planning, environmental monitoring and disaster management. However, Multimodal Large Language Models exhibit critical deficiencies in fine-grained spatial understanding within Remote Sensing, primarily due to a reliance on limited or repurposed legacy datasets. To bridge this gap, we introduce a large-scale dataset grounded in verifiable cadastral vector data, comprising 3.8 million annotated objects across 510k high-resolution images with 135 granular semantic categories. We validate this resource through a comprehensive instruction-tuning benchmark spanning seven spatial reasoning tasks. Our evaluation establishes a robust baseline using a standard LLaVA architecture. We show that while current RS-specialized and commercial models (e.g., Gemini) struggle in zero-shot settings, high-fidelity supervision effectively bridges this gap, enabling standard architectures to master fine-grained spatial grounding without complex architectural modifications.