SOGS SH Issue

[lucasoptml]

I think there is an issue in the SOGS loader or renderer causing some rendering artifacts on SH0? N?.

I'm seeing directional 'fog' on the SOGS, but it looks great using ksplat.

I also verified that it looks good on the playcanvas/react/blocks splat viewer.

pinecone-sogs.zip

I'm using the latest playcanvas/sogs python to convert this over.

[lucasoptml]

I think it might be because it is sh2 but it looks written for sh3 only?

[lucasoptml]

I confirmed it, found a fix, and submitted a PR for sh2, sh1 sogs support.

[61cygni]

Thank you!!

[lucasoptml]

I think it's doing it a little still, but it's a little foggy on the playcanvas viewer too, appears (same-ish), so I think spark is good now? not sure.

Playcanvas (gosh they use some kind of harsh color correction)

[asundqui]

I think it's doing it a little still, but it's a little foggy on the playcanvas viewer too, appears (same-ish), so I think spark is good now? not sure.

Playcanvas (gosh they use some kind of harsh color correction)

@lucasoptml I don't have the pinecone file as a PLY so I can't see for myself in other viewers, but do you expect the colors look more like the Spark screenshots above, or the PlayCanvas one? The way PC stores their RGB coefficients allows for hyper- (and hypo-) saturated colors, so I'm wondering if this harsh color correction is what you expect, or should it be more like Spark's color scheme?

[lucasoptml]

I would think it should just match the ksplat like spark and mkkellogg. When I look at the source photos the ksplat/ply spark looks really close to the images, and PC viewer looks way different.

It looks like it's maybe happening in just dark areas which is maybe gamma related? It seems view dependant so I think it's in sh1 or sh2. The conversion code looks simple though I have not tried verifying the correctness yet.
....

I noticed though that SOGS doesn't have enough precision (assume it is 16bit on the means or scale) for our use case, and I was getting flickering. I heard PC was looking into doing x splat bunches like ksplat to fix this. If I have time I may roll my own hybrid approach...

[dmarcos]

@lucasoptml how is spark 0.1.6 looking for you? is this fixed? thanks

[dmarcos]

@lucasoptml let me know if we can close this or should keep it open. thanks

[lucasoptml]

@lucasoptml let me know if we can close this or should keep it open. thanks

I think it is fixed now, and the conversion code looks good to me for sh1 sh2 sh3. Thanks

[dmarcos]

Thanks so much for verifying. Fixed by #109