Help me fix my vorobroken

[crude kayak]

Sry, I'm trying to understand what's wrong with my voronoi, I managed to narrow it down and I'm quite sure it's a stupid error somewhere. The issue lies somewhere in how I get points distance.
For each pixel I wish to have the position of the first and second closest points in the point cloud. Any other math comes from there.
The code is entirely C++ since I'm using cuda. My "vecNT" classes are similar to GLSL vec2, vec3, uvec2, uvec3 etcc
All the relevant math is done in this class

If I remove the randomization, it's always the bottom-right quarter of a point that's wrong, and I'm not understanding how/why

randomized

unrandomized

[odd anchor]

For my voronoi it didnt take nearly that much code

you can start by returning a sqrt value if your recieving end isnt already doing that

and if its 3D voronoi noise, i recommend having a starting distance of a large number

and if you need, i can send some code, its your choice on what type of code i send, one that makes sense and is in python, c++ or the hellish attempt on scratch

[ebon juniper]

ah I see, the randomized points are centered on the corner of the tile
so you need to subtract 0.5 to it so that its centered on the tile

or actually you can just make your random offset be in the range [0, 1] instead of [-0.5, 0.5], same thing

[crude kayak]

i'll gladly take a c++ example

but logically is the reasoning right? look at the points in the nearby "regions" and take the 2 closest?

[odd anchor]

Just 1

[odd anchor]

I meant to reply to this

But you just want to take the closest 1 point

Actually, I can just use pseudo code or smth

That's easier to do rn

[crude kayak]

see that's not enough, I don't want just the distance from the point, i want the distance from the edge too

and to know that I need the distance from the 2nd closest point, don't i?

[odd anchor]

I'm so confused on what you're trying to do...

[crude kayak]

for a given pixel know how far it is from the center of its cell and the edge

[odd anchor]

My voronoi noise is a grid of points which are translated by random numbers and then the distance is calculated between the nearest 27 points

[crude kayak]

that's what I'm doing in my code though

i'm just taking the 2nd closest too because I can't know the edge distance without it

[odd anchor]

I'm just confused by what you mean edge

[crude kayak]

i want orange and yellow, not just yellow
just realized my math for orange was wrong regardless

[odd anchor]

I feel like the edge is calculated by the average of those 2 points

Not the sample

But you can always have a random value corresponding to the closest point

[crude kayak]

yeah i don't mind it, just trying to understand why my corners act weird rn

the edge calculation is a "later" issue

[odd anchor]

Alright

Well, I have a grid of pixels let's say

And scale it to 0.01 or so to see the noise

Then I simply repeat for x, y, z values 3 times each

Based on the floor of the sample point

[crude kayak]

That's my
repeat_distance * (utils::math::floor(position / repeat_distance), where repeat_distance is your 0.01 and position is the sample point

[odd anchor]

Repeat distance?

[crude kayak]

the base distance of the points from each other (without randomization)

[odd anchor]

Hm

[crude kayak]

ok found out what was wrong: I assumed that the 2nd closest point must be within a 3x3 grid around the sample point. However in some cases it isn't.
Here the sampel point's (cyan) closest point is blue and second closest point (which i used to calculate distance from edge) isn't within a 3x3 grid

so... is there any better way to find the % of the sample point between cell center and edge?

cause scaled to 3 dimensions this goes from 3*3*3 checks to (3*3*3 /*core*/) + ((3*3 /*face "bulge"*/) * 6 /*faces*/)