Skip to content

Terrain generation tool, using real-earth data and deep learning techniques

Notifications You must be signed in to change notification settings

kenjinp/terrain-synth

Repository files navigation

Terrain Synth

Example-based Terrain Generation powered by real-earth datasets.

image of the terrain renderer

This is a platform for me to hack around on various machine-learning content generation / manipulation techniques. It uses the Hello Worlds library to render the terrain.

The current model is very naïve, expect the output to look crap.

Example Input

Terrain elevation data input
real earth terrain heightmap example used as input in GAN model

Example Output

(Silly) generated terrain from the pytorch GAN model
generated terrain heightmap from the GAN model]

Technologies

  • pytorch for ML learning
  • onnex runtime (https://onnxruntime.ai/) for model inference
  • react / typescript / vite for the app
  • Hello Worlds library for terrain rendering
  • react-three/fiber / threejs for 3D rendering
  • pulumi / aws for deployment

Datasets

Earth Terrain - From Kaggle user THOMAS PAPPAS