Interesting paper for someone fiddling with implicit surfaces, but it's gonna take a little while to digest it. Also there's no examples of resulting UV maps used in this (while on-model painting is simplified, afaik classically artists still preferred to get clearly delineated maps to modify in regular painting software even if stuff like Substance painter seems to have shifted that).
Implicit surface modelling tools are still evolving quickly, and one point that seems overlooked is how this would tie with authoring tools (since this is about a fairly raw numeric algorithm solution).
Post-modelling it should hold up, but would the maps retain similarity (or enable transferability) if the implicit model structure changes?
Imagine 2 disjoint spheres in space, the user drags them together, would this algorithm want to merge the UV mapping of them if they are smoothly joined? (no sharp creases).
Hi, I'm the first author of the article. Thank you very much for your interest in the method!
For the 2 spheres, it's as you wish ! It depends on where you place the uv system in the CSG tree, if it is above a single sphere then it will remain on it, if it's above the union, the field should extent on both.
The main interest of the method is the fact that the output remains implicit so one can easily plug it in a CSG pipeline.
Right, so in the case of the UV connecting to the 2 sphere's in the CSG if one uses a smoothunion function, would the UV islands join "from each direction" in a predictable fashion as the spheres overlap more and more?
Great timing on this paper. I actually just started tackling a problem that is essentially exactly what is under discussion here (creating a coherent UV set for implicit geometry), so I'm very looking forward to reading it in depth.
At a first glace through, it seems to be written at a good blend between concept and implementation followthrough, something that notoriously is not always there for CG papers :) And it's also refreshing to read something that is not neuro-AI-generation of this or that for a change!
Related question: I'm coding a demo (as in "demoscene") and want to displace a plane SDF (think e.g. a heightmap) for raymarching.
This produces some visible discontinuities in the plane at larger slopes. My surface is still at f(x,y,z) = 0 but the rest of the field is completely messed up. From the little maths I know, that is no longer an SDF because among other issues, it no longer satisfies the Eikonal equation mentioned in the article.
You need to take the minimum of the two SDF fields in the surrounding region (up to the global context if you need to) if you want it to maintain an SDF. The minimum of two SDFs is always a new SDF.
My suspicion is that the new plane isn't an SDF anymore, that is why you are getting artifacts or your ray marcher has problems with large scope angles, or something.
Interesting paper for someone fiddling with implicit surfaces, but it's gonna take a little while to digest it. Also there's no examples of resulting UV maps used in this (while on-model painting is simplified, afaik classically artists still preferred to get clearly delineated maps to modify in regular painting software even if stuff like Substance painter seems to have shifted that).
Implicit surface modelling tools are still evolving quickly, and one point that seems overlooked is how this would tie with authoring tools (since this is about a fairly raw numeric algorithm solution).
Post-modelling it should hold up, but would the maps retain similarity (or enable transferability) if the implicit model structure changes?
Imagine 2 disjoint spheres in space, the user drags them together, would this algorithm want to merge the UV mapping of them if they are smoothly joined? (no sharp creases).
Hi, I'm the first author of the article. Thank you very much for your interest in the method!
For the 2 spheres, it's as you wish ! It depends on where you place the uv system in the CSG tree, if it is above a single sphere then it will remain on it, if it's above the union, the field should extent on both.
The main interest of the method is the fact that the output remains implicit so one can easily plug it in a CSG pipeline.
Right, so in the case of the UV connecting to the 2 sphere's in the CSG if one uses a smoothunion function, would the UV islands join "from each direction" in a predictable fashion as the spheres overlap more and more?
If I understood what you mean, yes!
This example can be seen in the fast-forward: https://youtu.be/dK5JPHylzOg?si=zDHwaOL_xtx-_TJk
What a great paper. This paper is an example of why I find computer graphics to be the most interesting subset of computer science.
Hi, I'm the first author of the paper. Thank you so much for your comment, it means the world to me!
Great timing on this paper. I actually just started tackling a problem that is essentially exactly what is under discussion here (creating a coherent UV set for implicit geometry), so I'm very looking forward to reading it in depth.
At a first glace through, it seems to be written at a good blend between concept and implementation followthrough, something that notoriously is not always there for CG papers :) And it's also refreshing to read something that is not neuro-AI-generation of this or that for a change!
Related question: I'm coding a demo (as in "demoscene") and want to displace a plane SDF (think e.g. a heightmap) for raymarching.
This produces some visible discontinuities in the plane at larger slopes. My surface is still at f(x,y,z) = 0 but the rest of the field is completely messed up. From the little maths I know, that is no longer an SDF because among other issues, it no longer satisfies the Eikonal equation mentioned in the article.
Is there any way to fix this or am I SOL?
You need to take the minimum of the two SDF fields in the surrounding region (up to the global context if you need to) if you want it to maintain an SDF. The minimum of two SDFs is always a new SDF.
My suspicion is that the new plane isn't an SDF anymore, that is why you are getting artifacts or your ray marcher has problems with large scope angles, or something.
See "Deformations and distortions" paragraph in https://iquilezles.org/articles/distfunctions/ where it suggests playing with the ray-marching step size.
sdf(p + sin(p)*0.01)