LuxCoreRender Forums

Posted: **Thu Jun 11, 2020 1:13 am**

https://arxiv.org/pdf/2005.07547.pdf

This looks nice

https://twitter.com/jpantaleoni/status/ ... 5802252289

Posted: **Thu Jun 11, 2020 8:54 am**

Besides applying the same techniques to the handling ofparticipating media, a natural extension of this work wouldbe to combine it with adjoint-driven Russian roulette andsplitting [Vorba and Kˇriv ́anek 2016], which might help furtherreduce variance at a low additional cost. Another venue would be to couple it with progressive photon mapping [Hachisukaet al.2008; Hachisuka and Jensen 2009] or vertex merging techniques [Hachisuka et al.2012]. Yet another potential area is coupling it with the latest results on improved multipleimportance sampling [Ivo et al.2019; Karl ́ık et al.2019;Sbert and Elvira 2019]. Finally, it would be interesting toautomate the choice of the constantγin our biased estimatorto minimize total error, seen as the sum of bias and variance.

This part of the paper is really interesting.

Also the video by pantaleoni look like it can help to make realtime pathtracer by using stronger approximation reducing the required computation power at the cost of accuracy. At the begining i just tried to not confused it with demoising.

BTW the scene is really simple ( Area light / and the regular path tracer at the right is pretty fast too (strong Nvidia quadro ?) )

Posted: **Thu Jun 11, 2020 10:01 am**

At veeeeeeeeery first sight (after having reading the paper for 3 seconds), there is a little problem: it is based on the idea that light sources are intersectable (short version: it can not render a point light source, etc.).

There may be a workaround to the problem by considering any surface as a secondary light source of the original not-intersectable light source (i.e. adding a direct light sampling step) but there still tons of stuff it can not render if implemented as plain path tracer.

I have the feeling PhotonGI may be faster.

Posted: **Thu Jun 11, 2020 10:14 am**

I have the feeling PhotonGI may be faster.

you are probably very right photon Gi is a monster a real beast in term of speed and quality. the only small issue are about it is animation (with moving light and object).
precomputation also is negligeable in still but start to be a pain when you want to render animation as fast as possible. ahaha human always unsatisfied

LuxCoreRender Forums

Online path sampling control with progressive spatio-temporal filtering

Online path sampling control with progressive spatio-temporal filtering

Re: Online path sampling control with progressive spatio-temporal filtering

Re: Online path sampling control with progressive spatio-temporal filtering

Re: Online path sampling control with progressive spatio-temporal filtering