I need to keep track of these links
https://www.youtube.com/watch?v=7xL0udlhnqI
https://web.archive.org/web/20210925033 ... generation
Both contain valuable stuff about terrain generation and I keep losing one of those links
Terrain Texturing
-
FluffyFreak
- Posts: 1353
- Joined: Tue Jul 02, 2013 1:49 pm
- Location: Beeston, Nottinghamshire, GB
- Contact:
Re: Terrain Texturing
Putting a link here for future reference: https://github.com/kurtkuehnert/planeta ... n_renderer
The accompanying Master's thesis makes the case that the cubic spherical projection we use is the best of a bad lot, and proposes an extension to use the algebraic sigmoid function with the sigma parameter 0.87^2 to map UV coordinates to X/Y coordinates (pre-normalization I presume). The results shown look very promising, with a much more regular distribution of grid vertices. Still doesn't do anything about the square->triangle squash at the corners of each cube face though.
I've done a lot of thinking regarding sparse virtual texturing and other terrain geometry and projection methods, with icospheres being a primary one; icospheres have the singular problem of requiring equilateral triangles, which makes any kind of virtual / tiled terrain texturing very awkward as you have to find some sort of texture atlas addressing scheme that works with offset and flipped rows of triangles to have any kind of efficient use of texel storage.
In general, I think it's probably best to move forward with this extension to our existing cubic projection; it simplifies a lot of the complexity and allows virtual terrain tiles to be rendered and cached in a very straightforward manner.
The accompanying Master's thesis makes the case that the cubic spherical projection we use is the best of a bad lot, and proposes an extension to use the algebraic sigmoid function with the sigma parameter 0.87^2 to map UV coordinates to X/Y coordinates (pre-normalization I presume). The results shown look very promising, with a much more regular distribution of grid vertices. Still doesn't do anything about the square->triangle squash at the corners of each cube face though.
I've done a lot of thinking regarding sparse virtual texturing and other terrain geometry and projection methods, with icospheres being a primary one; icospheres have the singular problem of requiring equilateral triangles, which makes any kind of virtual / tiled terrain texturing very awkward as you have to find some sort of texture atlas addressing scheme that works with offset and flipped rows of triangles to have any kind of efficient use of texel storage.
In general, I think it's probably best to move forward with this extension to our existing cubic projection; it simplifies a lot of the complexity and allows virtual terrain tiles to be rendered and cached in a very straightforward manner.