claudaff/Cartographic-ControlNet

Generative AI in Map-Making: A Technical Exploration and Its Implications for Cartographers

This repository contains ControlNet models for generating accurate maps in controlled styles, as discussed in our paper: Generative AI in Map-Making: A Technical Exploration and Its Implications for Cartographers.

Traditional map-making often relies heavily on Geographic Information Systems (GIS), requiring domain expertise and being time-consuming. This work leverages recent advances in generative AI, specifically image diffusion models like ControlNet, to integrate vector data and guide map generation in various styles specified by textual prompts. This approach aims to automate and democratize the map-making process, enabling the creation of accurate maps with controlled spatial composition and semantic layout.

For more details, code, and further instructions, please visit the corresponding GitHub repository.

---

Models

Our four ControlNet models can be downloaded here.

1. Swisstopo.ckpt: Specialized model for Swisstopo style
2. OldNational.ckpt: Specialized model for Old National style
3. Siegfried.ckpt: Specialized model for Siegfried style
4. Combined.ckpt: Combined model, capable of generating map tiles in all three styles and used in our web app.

Class label	RGB color code	Swisstopo	Old National	Siegfried
Background	(255, 255, 255)	✓	✓	✓
Building	(82, 82, 82)	✓	✓	✓
Coordinate grid	(237, 240, 64)	✓	✓	✓
Railway (single track)	(219, 30, 42)	✓	✓	✓
Railway (multi track)	(144, 20, 28)	✓	✓	✓
Railway bridge	(226, 132, 115)	✓
Highway	(247, 128, 30)	✓	✓
Highway gallery	(231, 119, 28)	✓	✓	✓
Road	(149, 74, 162)	✓	✓	✓
Through road	(255, 103, 227)	✓
Connecting road	(128, 135, 37)	✓
Path	(0, 0, 0)	✓	✓	✓
Depth contour	(63, 96, 132)	✓	✓
River	(41, 163, 215)	✓	✓	✓
Lake	(55, 126, 184)	✓	✓	✓
Stream	(89, 180, 208)	✓	✓	✓
Tree	(63, 131, 55)	✓
Contour line	(164, 113, 88)	✓	✓
Forest	(77, 175, 74)	✓	✓	✓