Recommendations for changing the CMFS.

[mitsuba-renderer]

Recommendations for changing the CMFS.

[KelSolaar]

First light test:

The Google Colab notebook to generate the sensitivities data is here: https://colab.research.google.com/drive/18X8ICpZ6MZmWxOzXNpUF0ehb_emfUOf-

It makes me wonder if you have a build of the renderer (or recipe) that works on the Colab VMs? That could be useful! If not I will work on one when I have some spare cycles.

Note that I have an almost exactly 1 stop energy loss between the CMFS and the camera sensitivities that I need to investigate but it reminds me #14. The issue is most likely between the chair and the screen.

[KelSolaar]

Another related question is: Recommendations on how to output spectral-images? :)

[wjakob]

Discussing here is fine. But what is a "CMFS"? :-)

[KelSolaar]

Hi @wjakob,

First congratulations are in order, and I should have started there! Quite an awesome release, I'm super excited :)

CMFS stands for Colour Matching Functions, I would like to replace the CIE 1931 2 Degree Standard Observer data with cameras sensitivities.

I'm part of an Academy working group trying to come up with a gamut mapping algorithm for scene-referred data, and I would like to generate spectral imagery from multiple camera models.

I have been using existing hyper-spectral images and integrated them with narrow-band LEDs and various camera sensitivities here: https://academy-vwg-gm-hyper-spectral-images.imfast.io/ The problem with that approach (even though it generates offending data as we require) is that it does it too well, the entire image is affected! I would like to generate synthetic data in Mitsuba with mixed lighting and that is where the different sensitivities come to play.

Hope it makes sense!

Cheers,

Thomas

[wjakob]

Hi @KelSolaar,

that makes sense. As a quick-and-dirty solution, you could simply replace the CIE XYZ tables in src/libcore/spectrum.cpp by something else. A cleaner solution, and a technique for outputting arbitrary number of spetral channels should not be too hard by implementing a new meta-integrator that calls a nested "sub-integrator". This is already done in aov.cpp and and would just need some small changes to integrate against different spectral response functions.

Best,
Wenzel

[KelSolaar]

As a quick-and-dirty solution, you could simply replace the CIE XYZ tables in src/libcore/spectrum.cpp by something else.

Yeah that is what I have been doing for now (re-compiling is long though).

I reckon it would be 2 new useful features!

[Speierers]

It makes me wonder if you have a build of the renderer (or recipe) that works on the Colab VMs?

Not really. It would be great to have! Feel free to open a PR to add a recipe to the documentation, that would be really useful!

I am going to close this as this is not really an issue.