Perspective rendering 101?

Well, what you're talking about is "rendering". [smile] There's two major methods of rendering that you'll be interested in: Scanline-based rasterization (such as virtually all video games do) and raytracing (such as, well, raytracers do). There's also REYES, which luckily is utterly inappropriate for the sort of stuff you want so you can safely ignore it.

Each of these methods does perspective in a different way. The scanline rasterizer mathematically processes the vertices of the triangle using something called a "projection transformation" followed by something called a "perspective division" which get things looking correct. A raytracer, in contrast, simply sees what color of light is coming into the eye from each direction; the perspective this engenders is implicit in the directionality of the rays. Note that in both cases, the retina itself is not involved in "rendering perspective".

Anyways, if all you need is an input for your retina, what you need is just to render an image. Probably a small image. Note, however, that the consensus of the computer vision community is that just squirting an image into the business end of an ANN is not by itself a useful way of training a vision algorithm. Preprocessing is usually felt to be necessary to get useful results, regardless of the application.

Thanks, Sneftel, exactly what I needed to know. As for preprocessing, well, I'll jump off that bridge when I come to it, but the point of the project is for the NNs to evolve, so with luck they'll sort some of that out themselves. (We carbon-based types did! :) )