ive been trying to do a shader that would draw voronoi cell borders with constant width ( constant world space width).
my best result so far is by getting two closest point from point cloud(PC1 and PC2) getting the mid point between them(M). then doing a dot product between vector M-P and normalized PC1-PC2 and then thresholding that distance value to get the line. it works allmost beatifully. i get problems on the cell border crossings.
maybe my approach is completely off, any better thoughts?

cheers
Kustaa

Hey Kustaa,

I am hacking my way through Advanced Renderman and there is a paragraph describing the voronoi-noise including normalizing the veins. Have a look at this code (it's all in proper vex). You compile it to an otl with the command vcc -l MyVoronoi.otl voronoi.vfl

Cheers,
Hans

Haven't look at pagefan's file, yet.

There's also Stupid Renderman Trick for constant width noise lines:
Constant Width Shaders! by Mach Kobayashi from Pixar Animation Studios

A compiled Houdini version:
http://forums.odforce.net/index.php?/topic/6967-a-shader-each-day/page__view__findpost__p__46797 [forums.odforce.net]


http://renderman.org…mach2006.ppt.gz [renderman.org]

thanks for your replies Pagefan and rgd!

@Pagefan
i had a quick try with your constant width voronoi but i seem to get the same crossing problems as with my dot based solution. its more obvious with thicker lines.

@rgd
that derivative based “normalization” looks really intresting, i have to take a closer look at it. im afraid that it might have the same problems as the other two approaches. since the crossing problems appear where the pointcloud pair changes. maybe i need to read my points differently

i made again a picture that tries to explain what i think the problem is.
black lines are the cell borders, grey lines show where the pointcloud pairs change and white is the constant width border.

cheers!

Maybe trace the Voronoi, render it as lines with a width attribute and fix it in comp?

i think i got closer to the solution!

instead of considering just a pair of points from pointcloud i need to compare more of them. also it helped alot to write this in vex instead of trying to navigate in vop-wire-jungle.

this is my first shader written in vex an theres some issues that i need to clean up, especially with looping through the points, right now its “hardcoded” to go through 4 points. but i couldnt wait to show it.


surface
voro()
{
int loop;
int handle;
int success;
float data;
vector pc;
vector oP;
int iter;


oP = ptransform(“space:current”, “spacebject”, P);

handle = pcopen(“/home/vuoriku/Desktop/pc.pc”, “P”, oP, 9999, 4);

loop = 1;
iter = 0;
while( loop != 0 )
{

vector pc1;
vector pc2;
vector pc3;

success = pciterate(handle);

pcimport(handle, “P”, pc);

//gather first three points
if(iter == 0)
{
pc1 = pc;
}
if(iter == 1)
{
pc2 = pc;
}
if(iter == 2)
{
pc3 = pc;
}

//main loop
if(iter == 3)
{
vector m1 = (pc1+pc2)*0.5;
vector pm = oP-m1;
float dist1 = abs(dot(normalize(pc1-pc2), pm));

m1 = (pc1+pc3)*0.5;
pm = oP-m1;
float dist2 = abs(dot(normalize(pc1-pc3), pm));

m1 = (pc1+pc)*0.5;
pm = oP-m1;
float dist3 = abs(dot(normalize(pc1-pc), pm));

data = dist1;
if(dist2 < dist1 && dist2 < dist3){ data = dist2;}
if(dist3 < dist1 && dist3 < dist2){ data = dist3;}
}
loop = success;
iter += 1;
}
Cf = filterstep(0.03, data, “width”, 0.01);
}

nice!