Bridging Scales in Map Generation: A scale-aware cascaded generative mapping framework for seamless and consistent multi-scale cartographic representation

TL;DR

SCGM presents a scale-aware cascaded diffusion framework for multi-scale tile map generation from remote sensing imagery. It introduces a scale-modality encoding mechanism, a cascade reference strategy, and a dual-branch feature adaptation network to enforce cross-scale cartographic fidelity and seamless tile coherence. Evaluations on CSCMG and MLMG datasets demonstrate state-of-the-art performance across FID, PSNR, SSIM, and the Map Feature Perception (MFP) metric, with clear gains from cascade references and CLIP-based scale guidance. The approach advances real-time, semi-automatic cartography with potential applications in emergency mapping and urban planning, and the authors provide open-source code and data.

Abstract

Multi-scale tile maps are essential for geographic information services, serving as fundamental outcomes of surveying and cartographic workflows. While existing image generation networks can produce map-like outputs from remote sensing imagery, their emphasis on replicating texture rather than preserving geospatial features limits cartographic validity. Current approaches face two fundamental challenges: inadequate integration of cartographic generalization principles with dynamic multi-scale generation and spatial discontinuities arising from tile-wise generation. To address these limitations, we propose a scale-aware cartographic generation framework (SCGM) that leverages conditional guided diffusion and a multi-scale cascade architecture. The framework introduces three key innovations: a scale modality encoding mechanism to formalize map generalization relationships, a scale-driven conditional encoder for robust feature fusion, and a cascade reference mechanism ensuring cross-scale visual consistency. By hierarchically constraining large-scale map synthesis with small-scale structural priors, SCGM effectively mitigates edge artifacts while maintaining geographic fidelity. Comprehensive evaluations on cartographic benchmarks confirm the framework's ability to generate seamless multi-scale tile maps with enhanced spatial coherence and generalization-aware representation, demonstrating significant potential for emergency mapping and automated cartography applications.

Figures (9)

• Figure 1: The generation of multi-scale maps from remote-sensing images can leverage scale information and tile segmentation in multi-scale tile map samples.
• Figure 2: Overview of the SCGM framework: a hierarchical, self-cascading pipeline that progressively synthesizes multi-scale tile maps by refining smaller-scale outputs under explicit scale conditioning.
• Figure 3: The SCGM architecture leverages a VAE to transition the diffusion and reverse processes from pixel space to latent space. During training, the latent representation of the target map, $z_0$, undergoes a progressive transformation into $z_t$ through the diffusion process, followed by denoising achieved via a U-Net network. Two bespoke modules are introduced to further refine the denoising process: MFEncoder and SFAdapter. The MFEncoder integrates information derived from remote sensing imagery and cascading references to construct the conditional feature, $F_{\mathrm{cond}}$. Expanding on $F_{\mathrm{cond}}$, the SFAdapter produces multi-scale features that are subsequently merged with the outputs of corresponding U-Net layers through element-wise addition.
• Figure 4: Detailed structure of the MFEncoder, SPADE module and basic blocks.
• Figure 5: Examples of RS-Map tile sample pairs from scales 1:35,000 to 1:2,000 in the CSCMG dataset are provided, covers a broad range of cross-scale scenarios.