I was reading about procedural modeling and it made me wonder if anyone has tried getting the data of loads of different 3D models and training a generative classifier on them - so classify all the 3D models that are of chairs, all that are of swords etc.
Then by generating synthetic data one can generate more 3D models of objects without the need to create them by hand.
This could help a lot for CGI/video games etc. where typically one wants a large number of slightly different models to give more variety but artist time is very expensive.
I searched but couldn't find anything, but this seems way too simple to not have been done so I can only presume it is a bad idea for some reason?
[link][comment]