Sat2Flow: A Structure-Aware Diffusion Framework for Human Flow Generation from Satellite Imagery

TL;DR

Sat2Flow addresses the need for scalable origin-destination flow generation in data-scarce urban environments by relying solely on satellite imagery. It introduces a three-stage framework: satellite imagery encoding, a modality-joint contrastive module with a multi-kernel encoder, and a permutation-aware conditional latent diffusion module with permutation embeddings and Pi-Net; these components enforce structural coherence under regional index reordering. The method achieves state-of-the-art accuracy and robust distributional fidelity on 3,333 U.S. urban areas, outperforming physics-based and data-driven baselines and maintaining performance under permutation. This work enables globally scalable mobility modeling without region-specific auxiliary data.

Abstract

Origin-Destination (OD) flow matrices are critical for urban mobility analysis, supporting traffic forecasting, infrastructure planning, and policy design. Existing methods face two key limitations: (1) reliance on costly auxiliary features (e.g., Points of Interest, socioeconomic statistics) with limited spatial coverage, and (2) fragility to spatial topology changes, where reordering urban regions disrupts the structural coherence of generated flows. We propose Sat2Flow, a structure-aware diffusion framework that generates structurally coherent OD flows using only satellite imagery. Our approach employs a multi-kernel encoder to capture diverse regional interactions and a permutation-aware diffusion process that maintains consistency across regional orderings. Through joint contrastive training linking satellite features with OD patterns and equivariant diffusion training enforcing structural invariance, Sat2Flow ensures topological robustness under arbitrary regional reindexing. Experiments on real-world datasets show that Sat2Flow outperforms physics-based and data-driven baselines in accuracy while preserving flow distributions and spatial structures under index permutations. Sat2Flow offers a globally scalable solution for OD flow generation in data-scarce environments, eliminating region-specific auxiliary data dependencies while maintaining structural robustness for reliable mobility modeling.

Figures (5)

• Figure 1: Structural consistency in urban OD flows. Geometric transformations (a) and index permutations (b) preserve isomorphic spatial relationships, maintaining structural consistency under regional reindexing.
• Figure 2: Overview of the Sat2Flow framework, featuring three sequential stages (satellite encoding, multi-kernel contrastive learning, and diffusion generation), with the latter two stages integrated within the Modality-Joint Contrastive Module and Conditional Latent Diffusion Module, respectively.
• Figure 3: Pi-Net Architecture: Inverted U-Net Design for OD Flow Generation.
• Figure 4: Ablation study evaluating component contributions through CPC, RMSE, and NRMSE metrics.
• Figure 5: Case study analyzing model behavior across selected cities with diverse sizes.