The screenshot is from Graphite, yes. But I realized that it was not configured to use my Nvidia's GPU, but only the motherboard's chipset 😅 That's my bad.

Now it works way faster! (around 5 fps, but it's manageable)

Nice! When I saw the whole country was scanned, I had to look at what the Matterhorn looked like and I was not dissapointed (although I cooked my CPU to render ~50M points. I guess @brunolevy01.bsky.social 's graphite is not designed for this use case 😅)

May I also ask what's wrong with the SDF -> Marching cubes -> Remeshing approach?

Hello :)

Sounds like a cool surface reconstruction project!
Here is a link to the data in .ply : filesender.belnet.be?s=download&t...

with z being 1.5*x.

And here is how it looks in @brunolevy01.bsky.social 's software

Pointing out a common misconception (that I also used to be guilty of): min(s1,s2) and max(s1,s2) are **not** signed distance functions, but lower/upper bounds on the true SDF of the intersection/union.

iquilezles.org/articles/int...
dl.acm.org/doi/fullHtml...