https://cseweb.ucsd.edu/~ravir/Overbeck2009AWR.pdf
ok that’s weird shit i’m done with scienceposting for the day
they are doing something like raytracing to wavelets instead of pixels and then exploiting ?math? to get high quality images out of 32 samples per "pixel" but it gives you the same result as if you did 512 samples per "pixel" with traditional rendering systems.
so basically it just shits out a jpeg2000 directly and then unpacks that to give you the image, and this is sixteen times faster.
or something.
so basically it just shits out a jpeg2000 directly and then unpacks that to give you the image, and this is sixteen times faster.
or something.
Follow
💀
@icedquinn
My only question for stuff like this is:
is it actually accurate and "lossless" (in that it will give you the exact same result each time you run it);
or is it a speed hack like Nvidia DLSS, that is not actually completely accurate, and it's a "lossy" guessing game of what the image should mostly look like if it were the real thing.
