TopoVision
Extracting topographical data from maps using computer vision
Problem & Impact
Current architectural and civil engineering workflows face a significant bottleneck: converting 2D topographical maps (contour lines) into usable 3D digital assets. This process often requires tedious manual data entry or fragile procedural generation tools, slowing down site analysis and environmental simulations. TopoVision solves this by automating the translation of static height curve maps into dynamic 3D meshes. For Architects/Engineers: drastic reduction in hours spent digitizing land plots. For ReGen Villages: accelerates the work of designing self-sustaining communities by allowing rapid analysis of different land plots for flood risks, solar exposure, and terraforming needs.
Tech Stack
Key Features
- Computer Vision Pipeline: Utilizes deep learning (PyTorch) to interpret height curves and visual features from standard image inputs (JPG/PNG).
- Synthetic Training Pipeline: Overcomes data scarcity by leveraging a custom-built synthetic map generator to train the model on diverse topographical scenarios.
- 2D-to-3D Reconstruction: Directly outputs a 3D elevation model (Digital Elevation Model or Mesh) from a single 2D input image.
Screenshots
TopoVision pipeline demonstrating how we go from 2D image to 3D model
Timeline
Start Date
November 2025
Current Phase
Completed in February 2026
Affiliations
ReGen Villages
Maintenance & Deployment
Research project completed by KTH AI Society team in collaboration with ReGen Villages
Contributors
Mattias Kvist
Research Lead
Erik Lidman Hillbom
Researcher
Edoardo de Cal
Researcher
Annysia Dupaya
Researcher