The idea is, as opposed to combining the Half Vector Space perturbation with Veach's original perturbations,
instead using the Multiplexed Metropolis Light Transport(MMLT) mutations in order to more robustly handle glossy corners and the like,
while reserving the costly Half Vector Space perturbation only for paths where it is necessary such as Specular Diffuse Specular(SDS) and the like.
What i would like input on, would be a path difficulty metric to determine probability of using ether MMLT mutations or Half Vector Space perturbations,
based on the frequency of the interactions, the order they come in as well as the size of both the aperture and the light source.
As a side note i would like input on a heuristic for determining for both mutations whether to mutate all vertices or keep a section of the path fixed.
update: I have figured out that the idea of the techique index in MMLT which used to choose how much of of the path originates from camera as opposed to the light source and vice versa can be generalized to how much of the path is fixed as opposed to not being fixed. However instead of just choosing how much is fixed compared to being not so by accceptance and rejection it would be chosen like a step size in adaptive markov chain monte carlo where being mostly fixed is similar to a small step size while being barely fixed is like a large step size. I would like to also have a way to ensure that the vertices that get fixed will eventualy get mutated to ensure ergodicity of the Markov Chain, I have also figured out that fixing the vertices is really just a poor man's substitute
to just having small step sizes for those vertices com pared to the others.
Practical and theoretical implementation discussion.
1 post • Page 1 of 1
Who is online
Users browsing this forum: Google [Bot] and 4 guests