I realize that all I'm defining is the size of the texture atlas.
That's not what I was asking.
For example, take the "Building" dataset. There are 50 images, each image being 12.7 megapixels.
Just because I can set a texture atlas of 50 x 16k images, doesn't mean that I'll get that much resolution out of the source data. The source data has much much less info in those images.
So beyond a certain point, increasing the texture atlas any further just wastes space.
I'm just looking for a way to find the optimum texture atlas that projects the original images pixel per pixel (or as close to that as possible) onto the resulting texture.
Is there any way to do that?