search

rgl-epfl / unbiased-inverse-volume-rendering

Code for the paper "Unbiased Inverse Volume Rendering with Differential Trackers"
Other
54 stars 6 forks source link

Differential Residual Ratio Tracking #5

Closed andyyankai closed 1 week ago

andyyankai commented 1 week ago

Hi Merlin, Is Differential Residual Ratio Tracking used in this code base? I only see ratio tracking I think. Do you think it is easy to add Ratio Tracking or Residual Ratio Tracking on lastest mitsuba3? (i noticed you already implemented it in your project branch for mitsuba3 Thanks!

merlinND commented 1 week ago

Hello @andyyankai,

I believe that Differential Residual Ratio Tracking was not included in the main release Integrator because it did not significantly improve variance in our tests. However, the pseudocode is included in the paper (appendix B + listing 2) and the sampling method is implemented on the Mitsuba side (Medium::sample_interaction_drrt()).

In terms of updating the codebase to the latest Mitsuba 3 version, some of the following changes would need to be rebased : https://github.com/mitsuba-renderer/mitsuba3/compare/4b428a03ed4502725f1fca46499a136fbd9d8a95...unbiased-inverse-volume-rendering Note that some of these changes are just performance optimizations that could be omitted, or might have been implemented in the newest version already.

The volumetric path tracer should also be updated: https://github.com/rgl-epfl/unbiased-inverse-volume-rendering/blob/c80b375deca3197f89cdde3bc46b1630c4142c19/python/integrators/volpathsimple.py However, as a Python-based plugin, it should be relatively easy.

andyyankai commented 1 week ago

Thanks for the help!