Similar to This person does not exist or This artwork does not exist, how might I go about creating a This street map does not exist, including choosing an appropriate AI model and scoping features?
Challenges might include:
Whether it would be better to work from GIS data or from rendered map tiles. To work from tiles at the zoom level I would prefer, specifically street maps, I would need to set up my own local tile server per Open Street Map's website page on serving tiles.
Most of the world is sparsely populated but maps of imaginary cities would be more visually interesting.
a) This could be mitigated by only choosing cities and towns as training data. This might be harder to manage if I am training from GIS data rather than easy-to-curate map tiles. Or perhaps I would need to create a pre-AI project to have it identify cities generated from my tile server and only pass along city tiles, discarding tiles that don't have cities.
b) I can choose city segments and download data from Protomaps but it takes a few minutes to navigate to a city, choose a selection, and download the data. If I need to have 10 samples as training data, this is reasonable, but if I need 100 or 1000 then it would be too tedious. How would I determine how big a sample size would be needed?
At this point I'm not sure I'm asking the right questions. New to AI here.