NewtonMeshApproximateConvexDecomposition tests

[Bird]

I tried on a a variety of objects and it looks like there might be some problems. Here's the results http://www.hurleyworks.com/downloads/decomp_tests.jpg

I think Hans might be failing because the mesh is composed of disconnected pieces ...like body, hair, vest etc. Noodle has disconnected parts too. But the dragon and angel are scanned meshes and should have no disconnected pieces and I checked to make sure that is true. The concave-cube is just one piece too.

-Bird

[Julio Jerez]

Ye smay be some bug still, I has no test much.
Let us start from the simple ot the complex.
can you send me the The concave-cube is just one piece demo and the paramenters you are using for the decomposition?

then after that we try the more elaborated models. for that I wan to add teh mesh simplificaton fase.

I should be able to fix those bugs fially eassy.

[Bird]

Here's the OBJ : http://www.hurleyworks.com/downloads/concave_cube.obj

parameters:
maxConcavity = .2f
maxCount = 16

-Bird

[Julio Jerez]

I just run the concave test in the max plugin in and I had a pefect decomposition, and had two sides each matching the convex pieces of teh box.
I tryed with 0.2 tolerance and with 0.1 and butvbalues yeilded the same result.
I can do this, I will same teh result in the SDK as a test file, mayeb you can run it and see if taht is right.

one thing we need to do si to pu teh dScene format so that we can transfer data tah can be read exactly as it is by Max and bet lighwave
OBJ format max doe some scalling and it may be that way we see is no waht we get. the dScen fiel will at leat guarantte we are working wi the the same data.

I know there are bug still specially with concave pieces. this particalar model should no fail in any way.
can you send me the Handsolo model?

[Julio Jerez]

Ha I think I found the problem. It is no really a bug is related to file format and metric system.
when I load the model in Max internally is dimension less but when each file has a metric system scale.
when a model is loaded each plug-in in convert the model to a Max dimensions by multiply by the scale system. so your Box in file his about [0.5 x 0.5 x 0.5]
when I load the Box into Max, Max load the unit as they are in the file, and all processing is done on the file unit.
The reason Max and some other modeling packages do not apply the scale to the model is that if you do that many time,
then each time you save the file the file the new copy is slightly different doe to floating point rounding error creeping in.
when I call the plug in form Max I am also getting a this values

g default
v -0.5 -0.5 0.5
v -0.5 -0.5 -0.5
v -0.5 0.5 0.5
v -0.5 0.5 -0.5
v -0.2 -0.15 0.5
v -0.2 -0.15 -0.5
v 0.5 -0.5 0.5
v 0.5 -0.5 -0.5
and the function work fine with that specified tolerance.

However when I export the model, I do apply the metric scale to the model when in Max by default is a one max unit represent one inch. therefore my model is tiny. these are the values in meters.
<position float4="8" floats=
"0.012700 -0.012700 -0.012700 1.000000
0.012700 0.012700 -0.012700 1.000000
0.012700 -0.003810 -0.005080 1.000000
0.012700 -0.012700 0.012700 1.000000
-0.012700 -0.012700 -0.012700 1.000000
-0.012700 0.012700 -0.012700 1.000000
-0.012700 -0.012700 0.012700 1.000000
-0.012700 -0.003810 -0.005080 1.000000" />


when I call teh plugin form teh SDK demo wit teh experted model I get the same result, and teh eaosn is that teh tolarnce is too big for size model.

The telrace repsent the maximum distant form a concave face to teh surceav of teh Convex hull that surroudn the collsion of face that face beleng to.
an value of 0.2 is bigget than teh masimun size of the model so teh entien model is collapsed int one single conve hold.

I beleive lighway applis the scale to the model at loading time, and it work with real units as opposted to Max that is always dimension less.
you can verify this by setting a break point here. ..\coreLibrary_300\source\physics\dgMeshEffect3.cpp
line 370 dgMeshEffect::Iterator iter(mesh); and open the array
const dgBigVector* const points = mesh.m_points;
you will see that the point has small values (values in to meters)

if this is the case you can do few things.

-Scale the model before calling the function, and scale back after the call
-scale the tolerance (0.2 * 0.0254)
-see if the OBJ plug in has a Unit system so when you load the model it do not apply the scale.

On my side this weekend I will complete some of the missing features of the functions (mesh simplifcation, and dealling with deep cancavities, dealing with disjoint pieces)
also a better algorith for selecting the best convex cluster.

the after that I will see if I can change the concaviy to be a normalize value so that people will not have to worry about scale systems.

please see if my asumpsions are right

also do not forget to give me teh hans solo model. (OBJ is fine at this time)

[Bird]

I beleive lighway applis the scale to the model at loading time, and it work with real units as opposted to Max that is always dimension less.
you can verify this by setting a break point here. ..\coreLibrary_300\source\physics\dgMeshEffect3.cpp
line 370 dgMeshEffect::Iterator iter(mesh); and open the array
const dgBigVector* const points = mesh.m_points;
you will see that the point has small values (values in to meters)

Yes, LightWave works in real units.
I still see the original size when I break in dgMeshEffect: http://www.hurleyworks.com/downloads/debugMesh.jpg

Heres' Mr.Solo http://www.hurleyworks.com/downloads/Han_Solo.zip

I'll try scaling the model down to see if it works.

-Bird

[Julio Jerez]

if the NetwonMesh is geting the original Size in the model then it spould produce the correct partition.

Please try again after to dwonlaod the mates version.

If it fail we will have to some how save that data to ithe dScene Mesh, to make this a litle simpler, can you do that? it soe be no more that few houer of work.

[Bird]

if the NetwonMesh is geting the original Size in the model then it spould produce the correct partition.

Please try again after to dwonlaod the mates version.

If it fail we will have to some how save that data to ithe dScene Mesh, to make this a litle simpler, can you do that? it soe be no more that few houer of work.

I tried again with the latest svn 1166 and it's sill not working. I tried with higher and lower values for maxConcavity but it still just made 1 pieces.
I'll try to save the data to dScene Mesh, but for now, here's the vertex and indices in case that will work for you.

-Bird

Code: Select all

static const float vertices[] = {
   -0.5,    -0.5,    -0.5,
   -0.5,    -0.5,    0.5,
   -0.5,    0.5,    0.5,
   -0.5,    0.5,    -0.5,
   0.5,    -0.5,    -0.5,
   0.5,    -0.5,    0.5,
   -0.2,    -0.15,    0.5,
   -0.2,    -0.15,    -0.5
};

static const unsigned int indices[12][3] = {
   {3,1,0},
   {3,2,1},
   {1,4,0},
   {1,5,4},
   {6,5,1},
   {2,6,1},
   {2,3,7},
   {6,2,7},
   {7,3,0},
   {4,7,0},
   {5,7,4},
   {6,7,5}
};

[Julio Jerez]

It does not make sence, I did run the same values by freezing the scale in Max and saving the file as it, and it generated a correct partition. two convexes one for each wing.

if you can convert the data to a scene Mesh, it will be eassy to place mode itself in the SDK demo, and see what is wrong.
there must be something missing between what you are doing and what I am doing some where.

[Bird]

It does not make sence, I did run the same values by freezing the scale in Max and saving the file as it, and it generated a correct partition. two convexes one for each wing.

if you can convert the data to a scene Mesh, it will be eassy to place mode itself in the SDK demo, and see what is wrong.
there must be something missing between what you are doing and what I am doing some where.

Heh, probably something stupid I'm doing. I'm not quite sure what you mean by "scene Mesh". I have written code that converts my mesh data to a NewtonBody .... what else do I need to do?

-Bird

[Julio Jerez]

it is the dScene, it is a file format.

[Bird]

I'm getting a crash now in svn 1166 when I try to create CollisionCompound for the convex pieces.

It's crashing at line 373 of dgCollisionCompound.cpp because the dgNodeBase* const sourceNode is a NULL pointer.

I tried the NewtonDemos->CompoundColliison and it works ok so I assume there's something wrong on my end but I can't find it. Any ideas why dgNodeBase* const sourceNode would be NULL there?

-Bird

[Julio Jerez]

I'm getting a crash now in svn 1166 when I try to create CollisionCompound for the convex pieces.

It's crashing at line 373 of dgCollisionCompound.cpp because the dgNodeBase* const sourceNode is a NULL pointer.

I tried the NewtonDemos->CompoundColliison and it works ok so I assume there's something wrong on my end but I can't find it. Any ideas why dgNodeBase* const sourceNode would be NULL there?

-Bird

what is cratching? you use of the compound collision?

[Bird]

what is cratching? you use of the compound collision?

Yes, but it only crashes after I make a NewtonMeshApproximateConvexDecomposition. I can create a compound collision in other ways and there's no crash.

And your SDKDemo example with the chairs works... so it must be something on my end.

-Bird

[Bird]

If it fail we will have to some how save that data to ithe dScene Mesh, to make this a litle simpler, can you do that? it soe be no more that few houer of work.

I can save the data to dScene but am still hitting an _ASSERTE (0) on line 47 of dCollisionCompoundNodeInfo.cpp, so I can't generate a scene file if the collision type is CollisionCompound. I'm trying in the latest svn 1170.

-Bird