Planetmap presently uses a traditional cubemap projection to distribute information over a radial domain. This is effective, but leads to some areas having ~50% more detail than others. Adopting a nonlinear projection within each cube face would allow us to drastically improve that for a small additional sampling cost. A variety of such mappings are compared in Cube-to-sphere Projections for Procedural Texturing and Beyond, which ultimately recommends a 5th-order polynomial approach that ~doubles the (very small) cost of applying the projection in exchange for reducing maximum area distortion to 7%. The increased sampling cost is further offset by the capacity to significantly reduce the number of samples (e.g. vertexes) necessary to achieve a certain minimum level of detail.
Applying this would lead to improved storage efficiency and isotropy, and make it much easier to define isotropic functions (e.g. for tree placement) in terms of individual chunks.
Planetmap presently uses a traditional cubemap projection to distribute information over a radial domain. This is effective, but leads to some areas having ~50% more detail than others. Adopting a nonlinear projection within each cube face would allow us to drastically improve that for a small additional sampling cost. A variety of such mappings are compared in Cube-to-sphere Projections for Procedural Texturing and Beyond, which ultimately recommends a 5th-order polynomial approach that ~doubles the (very small) cost of applying the projection in exchange for reducing maximum area distortion to 7%. The increased sampling cost is further offset by the capacity to significantly reduce the number of samples (e.g. vertexes) necessary to achieve a certain minimum level of detail.
Applying this would lead to improved storage efficiency and isotropy, and make it much easier to define isotropic functions (e.g. for tree placement) in terms of individual chunks.