Support for different CFA patterns

[tvercaut]

Thanks for the nice work. It looks like the assumption of a standard 2x2 Bayer filter is assumed. Are there any plans to relax that assumption to work with other configurations such as Quad-CFA?

Some mobile phones (e.g. Xiaomi Mi 10 Ultra) are now equipped with such sensors.

Single samples (no burst) can be found here: https://github.com/darktable-org/rawspeed/issues/256 https://drive.google.com/drive/folders/1EPZEpvBHupHehsYI600f9cS7m_MaVhSQ?usp=sharing

[Jamy-L]

Hi and thank you for your interest in the project. Adding other array configurations is not planned because I think it would make the code less readable and would diverge too much from the paper.

However it is totally do-able and I encourage you to propose a fork if you feel like it may be useful. The key to make it work are :

• In merge.py, one should change the 2 in local_CFA[pixel_idy%2, pixel_idx%2] by the size of the pattern (4 for quad CFA).
• The function compute_grey_image in utils_image.py should still work when using the FFT because it does not uses the pattern, but I have no idea of how good the obtained grey image will look. In decimation mode, the way to obtain a grey image must be modified to work with the new pattern
• compute_guide_image in robustness.py should be adapted. We currently get the rgb image by decimating bayer quads and averaging the greens, but with a different pattern it should be different.

In the end, it all boils down to figuring out how to decimate a raw image to grey or to rgb with the new pattern. The easiest may be nearest neighbour intepolation for each color channel separately, then averaging to grey.

[tvercaut]

Many thanks for the fast response. As there are no current plans for it, I'll close this issue but would re-open it if/when we have any progress to share on our side.