[Bobboau]

I don't suppose you could give me that COB?
it might be my geometry filter if there is some weird geometry there.

[Kazan]

you answered your own question

[chief1983]

PM sent.

[Bobboau]

ok, I got it, it does have some odd geometry, got a bunch of holes, I'm thinking conversion to cob did this mostly though. I turned the geometry filter off and it seemed to get worse, unfortunately I just ran out of time so I can't work on this right away, but the geometry filter should be correcting these issues (if they are what I think they are) I'll have to work on it later tonight after I get back from my meeting.

[Kazan]

he probably has convex polygons.. convex polygons are not allowed

[Bobboau]

he does, en mass, but that is exactly what the geometry filter is suposed to fix, and it's not, this is very good for me, cause it means I've got something to work on tonight, but I wonder what's going on there, I think it's more than just convex polygons.

[Kazan]

how exactly do you make convex polygons out of concave ones...

[Bobboau]

simple, you find the convex points, that would be the ones who's internal angle (angle measured from the inside) would be more than 180, you can find them easily enough by taking the relative crossproduct of the two neighboring points and doting it with the poly's normal, anything <= 0.0 is a convex point, if you have more than one then you divide the polygon into two separate polygons by connecting the first and second convex point, as a fall back you start trying to make good polygons starting at the furthest point from the convex one, you test every poly you make to ensure it is not convex it'self and eventually you will find a way of splitting the polygon that will not result any convex polys. or at least that's what I reckon. I did act for people to try and find a way to break the geometry filter when I was making  it, someone finally did. (maybe)


ah, I see, my method fails if two convex points are right next to each other

[chief1983]

Well if it did, I guess I'm glad, but I can't take credit for the model, it was a former SWC member who left it to us to texture.

[Bobboau]

I think I just figured out a better algorithm, once you find a convex point, you reverse directions and look for the first point that will result in a convex point if split between it and the convex point you found split between the identified convect point and the one before the one that will cause a bad split, then recurse the remaining half of the polygon.

[Kazan]

hey bob... here is what i was thinking for the smoothing-reverse-engineer
it runs on a vector of pcs_polygons

Code: [Select]

iterate over all polygons
 for each polygon
    (return when condition met) 
    1st) check all points to see if they're faceted
    2nd) check all points to see if they're full smooth
    3rd) iterate over all points, find their neighbors, determine minimum and maximum smoothing angles which result
            in the correct resultant
           find the minimum angle valid for all points

sound good for you?

[Bobboau]

sounds like it should work, though I think there will probably be a lot of situations where it will produce a blob of full smooth with a bunch of different auto-faceted bits on the perimeter. it would probly be a good idea to allow the user some options, such as simply providing a global auto-facet angle if they know it only used one

[Kazan]

how would it do that? it's on a per-polygon basis

oh and we have a problem in the compiler... setting normnum to -1 (which means the norm wasn't found)... this shouldn't be possible...

[Bobboau]

well like I said, GLOBAL, if the user doesn't want to sit and wait for eternity for this to calculate they should have the option of using a less accurate method, such as just providing an angle to use on all polygons.

norm -1, I would bet rounding error is responsible for this, in stead of using vector==vector, you should use dot(vector,vector)>0.99f or something similar.

I have a geometry class in the works that is built around making a lot of these comparisons faster, it has a single list of points which is referenced by a list of verts which are used by a list of polygons, every point keeps a reference to every vertex which uses it, and every vertex keeps a reference to every polygon that uses it, it makes looking up stuff like neighbors extremely fast. it'l be (nearly) an N*logN operation rather than (worse than) N^2, because the largest chunck of data (points) will work on a ordered index, and you can use binary search on that. (keeping everything in order is the major coding problem of this method, especially given the interdependencies, but it's the only way to avoid an unacceptably long processing time). I have it set up so it can keep everything up to date all the time (this mode is mostly for geometry editing) or allow you to make a few changes and then fix it later.

[Kazan]

A.A should equal |A|*|A| not 1

A dot B = AxBx + AyBy + AzBz

but that would probably work.. problem is it would also produe false positives i think

[Bobboau]

well sence the normals are by definition suposed to be unit length, then it means the only variable involved is the angle between them, which should be close to 0 if they are about the same. and cos(0) == 1

as long as the test angle is less than what can be easily perceived then it shouldn't matter if it's a false positive.

[Kazan]

you're thinking A dot B = |A||B|cos(theta)

plus i would be altering operator==(vector3d, vector3d) so it needs to work for more than just norms

[Bobboau]

well, normals are used in very diferent contexts than most vectors, even though they are the same data type. I don't think you can make a universal solution for this, unless you make a vector3d derived class for normals.

[Bobboau]

ok folks new build in my sig, undo/redo functionality should be working.

please test and report any problems here and more importantly in mantis.

[Kazan]

awesomeness