How do the commercial renders perform antialiasing on one pass only? I see them shooting some photons (looks like big blocks), then blur the photons, then antialiase! I always though you had to render your image 4x+ the actual size!?
I'm not quite sure what you're describing, but the usual way to do antialiasing in a raytracing-based renderer is indeed to fire multiple rays per pixel (distributed over the pixel area), and filter.
A common optimization is adaptive supersampling, where the number of rays is adjusted for each pixel based on some error metric, e.g. local contrast. Pixels on or near geometry or shadow edges will get more samples, but pixels with nothing too interesting going on may just get one sample.
I see, so they only scan the edges and re-render them using more samples. But what about textures? If only one sample is used, wouldn't the textures also look bad?
Not necessarily - you can do nice texture filtering even with just one ray, by using mipmaps and anisotropic filtering, etc. in much the same way that GPUs do. This is faster than shooting multiple rays because you don't need to do so many intersection tests, and you can precompute at least part of the texture filtering (in the form of mipmaps).
oh ok, that make sense, thanks Reedbeta.