Just a general question, is it a technique to somehow memoize our L(p, omega_i) (maybe hashing close values of p and omegas to the same buckets) so we don't do repeated computation (like a dynamic programming type approach)? I guess since p and omega are real-valued, the space requirements for bucketing them maybe wouldn't be worth it?
Osoii
For simple scenes, if we still use only one ray sample for Monte Carlo rendering, will it cause problems? Since there's not much bounce of the ray, so the sample result may deviate from the expected result a lot(the variance is large)
keenan
@Alex Yep, there are a bunch of caching schemes; as you point out, since the parameters are real-valued, it's impossible to cache all the radiance values. Instead, you might do some kind of "density estimation," i.e., guess what the radiance might be by interpolating between nearby cached values. One classic technique is irradiance caching, which is nicely summarized in Wojciech Jarosz' thesis. Another classic is photon mapping, and there's a nice little book on how to implement your own photon mapper. (Maybe you can add one to Scotty 3D! :-D)
Just a general question, is it a technique to somehow memoize our L(p, omega_i) (maybe hashing close values of p and omegas to the same buckets) so we don't do repeated computation (like a dynamic programming type approach)? I guess since p and omega are real-valued, the space requirements for bucketing them maybe wouldn't be worth it?
For simple scenes, if we still use only one ray sample for Monte Carlo rendering, will it cause problems? Since there's not much bounce of the ray, so the sample result may deviate from the expected result a lot(the variance is large)
@Alex Yep, there are a bunch of caching schemes; as you point out, since the parameters are real-valued, it's impossible to cache all the radiance values. Instead, you might do some kind of "density estimation," i.e., guess what the radiance might be by interpolating between nearby cached values. One classic technique is irradiance caching, which is nicely summarized in Wojciech Jarosz' thesis. Another classic is photon mapping, and there's a nice little book on how to implement your own photon mapper. (Maybe you can add one to Scotty 3D! :-D)