Question about the Unbiased Inverse Volume Rendering with Differential Trackers paper

[leroyvn]

I've been trying to use the local majorant implementation developed for the Unbiased Inverse Volume Rendering with Differential Trackers paper. I put together an example relevant to my use case and observed artefacts when using the volpath integrator with a majorant grid. Are there issues or limitations I could have overlooked?

[merlinND]

Hello @leroyvn,

Have you checked the volpathsimple integrator used in the paper and the corresponding Mitsuba branch, e.g. medium.cpp?
The diff is probably a good place to check for small details.

[leroyvn]

Hi @merlinND, thanks for your reply. I'm using the latest commit on the unbiased-inverse-volume-rendering branch. I was aware that the volpathsimple integrator implements the methods presented in the paper, but I need surfaces in my scene.

So maybe my question should rather be: were the results produced by the volpath with and without a majorant grid compared already? (Happy to investigate and fix issues, I just want to be sure that the artefacts I'm seeing are not a known issue already.)

[merlinND]

I don't think that the majorant grid was tested & validated for integrators other than the one used in the paper (the Python-based volpathsimple integrator).

[leroyvn]

Pretty clear, thanks!