Discrepancy in SH coefficient computation between code and paper

[deivse]

Hi, looking at the implementation it looks like 0th order SH is initialized to the keypoint color in image A (current main image), and all other SH coefficients are zeroed, but in the latest version of the paper is it claimed that all SH coefficients are computed based on directions of the cameras. Are you planning to update the code? Did you find the approach in the current public code is equivalent to what's described in the preprint?

I'm currently working on integrating this into my pipeline for a paper. We are targeting ECCV26, so the deadline is very soon, and we would appreciate having some clarification on this before publishing. Additionally if there's some updated code we could get access to before you submission, that would help very much, especially if it handles images with different sizes. @dimakot55

In any case, I appreciate your work, and wishing you luck if you're also targeting ECCV26 or another conference!