MapGAN: An Intelligent Generation Model for Network Tile Maps.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Geographic Information Systems

Background:

• Generative Adversarial Networks (GANs) excel at image synthesis but often produce low-quality outputs with insufficient detail.
• Existing GAN-based image translation models struggle with the accuracy and aesthetic demands of electronic map generation.

Purpose of the Study:

• To introduce Map Generative Adversarial Networks (MapGAN), a novel model for accurate and efficient multitype electronic map generation.
• To enhance GAN performance for map generation by improving generator architecture and incorporating a classifier.

Main Methods:

• MapGAN builds upon the Pix2pixHD architecture, enhancing the generator and adding a classifier.
• The model learns distinct characteristics and style differences across various map types.
• Experiments were conducted using datasets from Google Maps, Baidu Maps, and Map World.

Main Results:

• MapGAN demonstrated superior performance in both one-to-one and one-to-many map generation tasks.
• Generated electronic maps exhibited optimal quality in terms of visual appeal and quantitative evaluation metrics.
• The model effectively captures and replicates the nuances of different map styles.

Conclusions:

• MapGAN offers a significant advancement in generating high-fidelity electronic maps.
• The proposed model addresses limitations of previous GANs in detail and quality for map generation.
• MapGAN provides a robust solution for creating accurate and aesthetically pleasing electronic maps.