I have dreamed about a pipeline that can take elevation data, building data, maps, and streetview imagery to process a city into all the components you would need to build a proper map for one of the GTA engines, including all the other objects like trees, fire hydrants, signs, light posts, traffic lights, etc.
The building textures could be recreated from the low-res streetview imagery using a pipeline to scale / relight them using image-to-image models. Building shapes could be rebuilt using generative models for 3d mesh components. Road & sidewalk quality (how clean, how faded lines are, etc) could be inferred from imagery as well to be applied to the map.
Super cool project, hoping it continues to evolve!
Finally got round to vibe coding something in Godot this summer for my city. It used GeoTIFFs on AWS S3 at s3://copernicus-dem-30m/ for elevation. No authentication required, each tile covers 1°×1°. Combined this with OpenStreetMap. Was enough to get started and satisfy curiosity. As you state, lots more that could be done.
I've long wanted to make a fun milsim game using 3D map and environmental data, pretty much since the moment I tried out Google Earth VR in 2019. I'm not passionate enough about it to do it on my own, but if anyone else ever gets taken by the idea and wants to make it happen, I'm game.
Eh, something with enough sliders to tweak the sim into other types of sandboxes seems most obvious
Then players can shape entity generation, gravity, to their vibe
Not quite Blender level turn teapot into a unintelligible smear of geometry range of tuning
The baseline being Earth world model then tweak entities individually, and such changes spread through the entity population over subsequent ticks of the world loop
Even include ability to speed up and slow the diffusion of the new parameters
GTA games require a lot of effort outside of this, but the concept of mass-producing a GTA style open world game for any given environment is very compelling. Probably frightening to Rockstar themselves but such is the course of technological development.
The building textures could be recreated from the low-res streetview imagery using a pipeline to scale / relight them using image-to-image models. Building shapes could be rebuilt using generative models for 3d mesh components. Road & sidewalk quality (how clean, how faded lines are, etc) could be inferred from imagery as well to be applied to the map.
Super cool project, hoping it continues to evolve!