>Our book is a thorough guide to Delaunay refinement algorithms that are mathematically guaranteed to generate meshes with high quality, including triangular meshes in the plane, tetrahedral volume meshes, and triangular surface meshes embedded in three dimensions.
While true, it also produces a topology that makes you want to throw your PC out the window.
However… if you combine this with other algorithms for decimation and approximate visual edge determinism, you can end up with pretty solid topography while still maintaining the details that delaunay gives you.
What’s really interesting is the work going on with gaussian splats, point clouds, retopology, and remeshing. Taking photos of a place and getting back 3D scenes complete with textures from your photos. Mmmmm. But it’s likely a matter of time before AI can do this more effectively.
Delone triangulation is awesome, don’t get me wrong, but you end up with triangles with odd lengths and vertices in weird places. Great for capturing approximation of detail, pack them verts in there!
Horrible for real-time applications. Often these meshes need to be simplified along the normals to reduce vertex count and produce a better topology (cleaner triangles that cover more space with less faces). The original meshes are still useful for normal baking.
Strictly speaking from a graphics perspective. It’s far more useful outside of this space.
One of the youtubers [0] I follow uses a Delaunay triangulation pattern on points that I _think_ are doing a random walk as their intro screen. There's something very satisfying about this pattern. I'm not sure how the choice of colour for the triangles is made, but I could watch it as a screen saver all day.
There could be some randomness to the points, but I definitely wouldn't call it a "random walk," which would have very jittery paths. These points are moving along some smooth curves.
I think it's probably computing a Delaunay triangulation and continuously updating it as the points move, but I didn't attempt to verify that.
The colors seem to be based on a simple color gradient going from top to bottom, but it's a bit more complex than that. The color inside each triangle is not constant. Interestingly the blue channel appears to be maxed out over the whole thing (not counting the vertices), which should simplify the analysis if you want to reverse engineer it, but I'll stop there.
While true, it also produces a topology that makes you want to throw your PC out the window.
However… if you combine this with other algorithms for decimation and approximate visual edge determinism, you can end up with pretty solid topography while still maintaining the details that delaunay gives you.
What’s really interesting is the work going on with gaussian splats, point clouds, retopology, and remeshing. Taking photos of a place and getting back 3D scenes complete with textures from your photos. Mmmmm. But it’s likely a matter of time before AI can do this more effectively.
Can you elaborate?
Horrible for real-time applications. Often these meshes need to be simplified along the normals to reduce vertex count and produce a better topology (cleaner triangles that cover more space with less faces). The original meshes are still useful for normal baking.
Strictly speaking from a graphics perspective. It’s far more useful outside of this space.
[0] Sebastian Lague - https://www.youtube.com/watch?v=vTMEdHcKgM4
I think it's probably computing a Delaunay triangulation and continuously updating it as the points move, but I didn't attempt to verify that.
The colors seem to be based on a simple color gradient going from top to bottom, but it's a bit more complex than that. The color inside each triangle is not constant. Interestingly the blue channel appears to be maxed out over the whole thing (not counting the vertices), which should simplify the analysis if you want to reverse engineer it, but I'll stop there.