HPix: Generating Vector Maps from Satellite Images

TL;DR

HPix tackles vector-map generation from satellite imagery by introducing a hierarchical GAN with global and local generators to capture coarse structure and fine detail. The global stage produces a coarse vector map, while the local stage refines it using the original image to reduce artifacts, driven by PatchGAN discriminators and an $L1$-regularized objective in the combined losses $\mathcal{L}_{global}^*(G, D_G)$ and $\mathcal{L}_{local}^*(H, D_H)$ with weight $\lambda$. On a maps dataset, HPix outperforms Pix2Pix, CycleGAN, MapGen-GAN, and CSCGAN, achieving pixel accuracy $61.04\%$, PSNR $26.98$, and SSIM $0.752$, with the local generator enhancing visual fidelity. The approach is demonstrated on road-intersection mapping and building-footprint clustering to yield interactive vector maps, underscoring practical utility for urban planning and disaster response.

Abstract

Vector maps find widespread utility across diverse domains due to their capacity to not only store but also represent discrete data boundaries such as building footprints, disaster impact analysis, digitization, urban planning, location points, transport links, and more. Although extensive research exists on identifying building footprints and road types from satellite imagery, the generation of vector maps from such imagery remains an area with limited exploration. Furthermore, conventional map generation techniques rely on labor-intensive manual feature extraction or rule-based approaches, which impose inherent limitations. To surmount these limitations, we propose a novel method called HPix, which utilizes modified Generative Adversarial Networks (GANs) to generate vector tile map from satellite images. HPix incorporates two hierarchical frameworks: one operating at the global level and the other at the local level, resulting in a comprehensive model. Through empirical evaluations, our proposed approach showcases its effectiveness in producing highly accurate and visually captivating vector tile maps derived from satellite images. We further extend our study's application to include mapping of road intersections and building footprints cluster based on their area.

Figures (11)

• Figure 1: HPix Architecture
• Figure 2: Architecture design of (a) global generator with nested skip connection network and its components: (b) encoder block, (c) decoder block, and (d) transition block.
• Figure 3: Architecture of local generator
• Figure 4: Architecture of (a) PatchGAN discriminator and (b) its CNN block
• Figure 5: Maps dataset sample showcasing some satellite images and their corresponding vector tile maps.