Sculpty - which pixels are used at each LOD

If you have for example a 32 x 32 sculpty texture, which pixels are used for the vertex positions at each level of detail?, ie which are the important key pixels?

Also what about in a 64 x 64 texture?

SculptySpace by Sculpty Carver is really good for previewing and LOD things. Also comes with a handy texture with LOD vertices marked with dots. It's even friendly with stitching and the various sculptie types.

Invaluable.

http://sculptyspace.com/

It is hard to 'write a discription' which vertices are key...

A 32x32 up close shows all detail or LOD level 3, you see 32x32. As you move out that halves to 16x16 and the next step out is level 1 at 8x8.

There is a tutorial on machinimatrix.org at: http://blog.machinimatrix.org/2008/06/04/precision_sculpties_lod/

While for Blender it shows the important vertices. You will see what is happening. But unless you have a program capabile of showing you Mutlirez levels, it will be hard to tell which are the important vertices.

https://wiki.secondlife.com/wiki/Sculpted_Prims:_Technical_Explanation#Rendering_in_Second_Life_viewer has a table of which pixels are sampled for a 64x64 texture. The different LODs are obtained by sampling in a similar but sparser pattern.

sculpted prims use 33x33 vertexes (it changed from version to version and is now at 33x33, but backwards compartible to all previous systems).

so, lets focus on the one and only good and efficient methods; 64x64 lossless, a chess pattern of 2x2 pixels that have the same color like
aabb
aabb
ccdd
ccdd
but shifted by one pixel, tiled, resulting in one more row and or one more collumn:
theres 4 ways, all work, the only difference is the mapping of the vertexes and their texture:

nowadays it always uses 33x33 vertexes on a scumlptmap (for TYPE PLANE), but colinear polygons and their vertexes on the same place are skipped (generally at polar regions), wich results in less vertexes.
Before losless compression got perfected some systems sometimes failed in precision while others did not.

aabb
aabb
ccdd
ccdd
results into 32x32 vertexes, because there are 2 vertexes at one of each border per row and column that have the same color/position and so they merge into oen vertex.
When used for spheres this will not have a central equator od vertexes but have face-centers along its equator.

aabb
ccdd
ccdd
eeff
results into 32x33 vertexes, most commonly used, because that results in a nice central 32-vertex-equator or TYPE_SPHERE or CYLINDER)

abbc
abbc
deef
deef
results into 33x32 vertexes, almost never used, makes no sense to use for any sculpt type, use another instead.

abbc
deef
deef
ghhi
results into 33x33 vertexes, new and most vertexes per image, use for all other types.

which ones are skipped:
on the texture (from above) there are 33x33 dots layd, sill not sure how exactly.
on every LOD reduction every other dot from the (previous LOD) is just skipped, but the first and last dot are always kept.

32 x 32 is only suitable for a torus sculptie type without losing detail. Every pixel relates to a vertex.

64 x 64 will do all the sculpt types. The key pixels are (numbering from bottom left as 0,0):

Torus: Even X, Even Y
=( 0,0; 0,2; 0,4; 2,0; 4,0; 2, 2 etc )
Cylinder: Even X, ( Even Y, 63 )
Plane: ( Even X, 63 ), ( Even Y, 63 )
Sphere: 32, 0; 32, 63; ( Even X > 1 ), ( Even Y > 1 )

From my own experiments it is (plane topology, unstitched)....
For 64x64 pixels, zero-based indexing...

Highest LOD : 0,2,4,6,8,....,58,60,62,63 = 33
Second LOD : 0,4,8,12,.... .,52,56,60,63 = 17
Third LOD : 0,8,16,24,32,40,48,56,63 = 9

For stitched topologies the last (63) is replaced by a duplicate of 0, to achieve stitching (x only for cylinder, both for torus). For the sphere, x is stitched, and in y, all the 0 and 63 are replaced by a single vertex from one pixel (32? I'm not sure which one). Also of note is that only the forward-upward diagonals are used to make triangles (I think).

I am not certain of any of this, but I put it here so that I can learn from others who may correct me.

Apologies... a greater authority preceded me while I was writing!

Well part from having two "Highest" LODs doesn't look bad

I find a simple guide is enough for UV Map editing in Blender. Just load into the UV Image Editor and enable "Snap to Pixels" in the UVs menu.

The guide file also shows what not to do with your LODs, just import it as a sculptie and change the multires level to see the mess you can get into It's best viewed from front or side.

Thanks everyone. There's a lot of usefull information on this thread.