Show HN: TerrainSR – fast, realistic heightmap upscaling model

Category: ai-ml

Tags: terrain-generation, heightmap-upscaling, deep-learning, gaming, simulation

Score: 7.3/10 (Innovation: 8, Technical: 7, Documentation: 7, Utility: 7)

TerrainSR is a deep learning model that upscales low-resolution terrain heightmaps (100m) to high-resolution (10m) while avoiding human-made artifacts like buildings and roads. It is fast enough for real-time use, generating a 50x50km patch in under a second on a modern GPU, making it suitable for games and simulations. The model is trained on high-quality elevation data and released with permissive licensing.

Target audience: game developers, simulation engineers, GIS professionals

Repository: https://huggingface.co/joe-gibbs/terrainsr

View on Hacker News