Scientists in the ML community always use real world pictures and find very hard to label data.
With video game engines we can generate infinite amounts of data, with pictures in different angles, light, shading, noise, shader effects, occlusion, transparency, ...
With shaders, it is also easy to generate pixel accurate labels for segmentation tasks, face recognition in scenes with people occluding each other, ...
Why don't we see ML scientists using Unity or Unreal Engine to create infinite amounts of data ?
[link][21 comments]