Looks dirty and with some artifacts compared to Supersplat

[heimeii]

spark (no-aa: 0.3, aa: 0, focalAdjustment: 2, premul alpha)

supersplat

Supersplat looks cleaner on the wall, which seems to be related to the color precision.

You can get the test file from
https://drive.google.com/file/d/1Ywt6RJ_OmBJCSReqJt5er9nPS-AVaBa_/view?usp=drive_link

Regards.

[heimeii]

It seems to be related to this change #134

I think the calculation of shn scale doesn't seem correct.

sh1 scale 63 -> 126
sh2 scale 0.5 -> 2
sh3 scale 31 -> 64

current:

fixed:

But it still looks a bit dirty

[mrxz]

It seems to be related to this change #134

I think the calculation of shn scale doesn't seem correct.

sh1 scale 63 -> 126
sh2 scale 0.5 -> 2
sh3 scale 31 -> 64

@heimeii Good catch, can confirm that the issue has to do with the spherical harmonics. For the sh3 scale it should be 62. If you want, feel free to open a PR with these changes.

With the above fixes the difference is indeed smaller, though some discolouration can still be seen with Spark:

This only seems to be an issue with the third degree spherical harmonics, as limiting it to 2 (maxSh = 2) gives the following:

[asundqui]

This is absolutely correct, our mistake in introducing this error! The final numbers 126, 2, and 62 look correct to me after going through the logic. Let me PR a fix real quick. Thank you for finding this and sorry for the regression!

[dmarcos]

Thanks. Fixed by #144

[dmarcos]

0.1.8 release with the fix is out now