Volumes Of Fun
http://www.volumesoffun.com/phpBB3/

3D surface reconstruction for volumetric voxel data
http://www.volumesoffun.com/phpBB3/viewtopic.php?f=2&t=11		 Page 3 of 7
Author:  David Williams [ Sun Feb 22, 2009 2:22 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Have you still got the line:

Code:
computeNormalsForVertices(&g_volData, *ispCurrent, SOBEL);


Do you still see the same image if you change between SIMPLE, CENTRAL_DIFFERENCE, and SOBEL? For me it did make a difference, but I am testing within the Thermite 3D engine not the PolyVox OpenGL example.

As you may remember, I don't have any lighting in my OpenGL example. I think it would be best if you send me your modified version of the example which does include lighting (I guess it's still one .cpp file?). Then I'll see if I can fix it.
Author:  Gajananan [ Sun Feb 22, 2009 2:38 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Quote:
computeNormalsForVertices(&g_volData, *ispCurrent, SOBEL);


Yes I had this code.

Quote:
Do you still see the same image if you change between SIMPLE, CENTRAL_DIFFERENCE, and SOBEL?

I could not see any difference.

Quote:
I think it would be best if you send me your modified version of the example


I have sent you the files.. from my gmail.. please check in case if it turns to be spam.
In the code I also had the code for modifying the voxels , after running it , just press 'd'... you will see the changes.

Thanking you for paying attention to the problem..

/Gajananan
Author:  David Williams [ Sun Feb 22, 2009 5:40 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Ok, I got your code. Actually there was some stuff missing (CVector3 and GLUI) but I just commented that out for the time being. I also went back to using createSphereInVolume() rather than LoadVerticesFromRawFile(), and removed some of the lights to make it easier to see what was going on.

I've emailed you the result, and when I run it I get the following image:Image

As you can see, things are much smoother (though not perfect...).

Try my new code. If you do not see the same image as me, then we have to try and debug. In this case, set a breakpoint or some kind of print statement to make sure that Line 32 of GradientEstimators.cpp is being executed.
Author:  Gajananan [ Sun Feb 22, 2009 6:31 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Thanks for your prompt reply.

Quote:
Actually there was some stuff missing (CVector3 and GLUI) but I just commented that out for the time being.

Sorry , I forgot to remove that code.


When I run it I get the following image:
Attachment:
spehre-feb23.JPG
spehre-feb23.JPG [ 41.51 KiB | Viewed 3051 times ]


Quote:
As you can see, things are much smoother (though not perfect...).

Yes I can see it in the image rendered in your version.

Quote:
Try my new code. If you do not see the same image as me, then we have to try and debug. In this case, set a breakpoint or some kind of print statement to make sure that Line 32 of GradientEstimators.cpp is being executed.

Yes . I try to put a break point at "computeNormalsForVertices" but I was never able to step into ... it always step over..

Then I put a print statement to make sure that Line 32 of GradientEstimators.cpp.. but print statement did not print...quite strange!

Note that I just use the same code you sent me in the mail after removing my old main.cpp code.


I will further look into it! not sure what went wrong in my case.
Author:  David Williams [ Sun Feb 22, 2009 7:06 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Try putting the following line before the call to 'computeNormalsForVertices()'

Code:
ispCurrent->m_v3dRegionPosition = regLowerCorner;
Author:  Gajananan [ Mon Feb 23, 2009 12:47 am ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Oh Great... Finally got the image as like yours.. thanking you very much for your great support.. I will work on real time modification of data as well.
Before trying to integrate polyvox into my code... just want to make sure what I should do.

- include header in PolyVoxCore
- include header in PolyVoxUtil
- refer PolyVoxCore.lib
- refer PolyVoxUtil.lib
- put PolyVoxCore.dll in dubug or release folder of my application
- put PolyVoxUtil.dll in dubug or release folder of my application

Is that enough?
Author:  Gajananan [ Mon Feb 23, 2009 2:32 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
I was able to refer polyvox in my application as I wrote before.. though extractReferenceSurface does not create 'vecVertices'.. means that this function is called ..seems not executed.. Do you think anything special setup with PolyVox to be used in other application?
Author:  David Williams [ Mon Feb 23, 2009 11:54 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Gajananan wrote:
Oh Great... Finally got the image as like yours.. thanking you very much for your great support.. I will work on real time modification of data as well.


Well that's strange, because the second fix was nearly the the same as the first one. Compare the first fix:

Code:
singleMaterialPatch->m_v3dRegionPosition = region.getLowerCorner();


To the second:

Code:
ispCurrent->m_v3dRegionPosition = regLowerCorner;


They are logically the same The first fix was placed right at the end of the extractReferenceSurface() function, and the second was placed immediatly after it. The only real difference is that the first fix went into the PolyVox library, whereas the second one went into your application code. Maybe you did not rebuild PolyVox properly, or copy the .dlls, after you applied the first fix?

And yes, as far as I can recall you just include the headers, link against the .lib, and then place the .dlls in the output directory (if that is not done automatically). If that doesn't work, try inserting some print statements or breakpoints?
Author:  Gajananan [ Tue Feb 24, 2009 3:52 am ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Quote:
Maybe you did not rebuild PolyVox properly, or copy the .dlls, after you applied the first fix?

I did compile the code and tested it with out the following code I don't get the result as yours
ispCurrent->m_v3dRegionPosition = regLowerCorner;

Quote:
And yes, as far as I can recall you just include the headers, link against the .lib, and then place the .dlls in the output directory (if that is not done automatically). If that doesn't work, try inserting some print statements or breakpoints?

Thank you for your advice. I just fixed the problem, rendering works fine.

Though I have an issue in data size. For 128 ^ 3 data size, I could not get the full object model. What I do now, i have 668-668-469 size data, then cropped it to be 128 ^ 3, this lead to whole in the surface. I am trying to load 256 ^ 3, it leads to memory problem. I have to tackle this problem as Polyvox limit the size to be power of 2.

Once I solved it I could think about how I can modify the voxel so that surface can be changed.
I still want to improve the smoothness. Because I am not sure how it would behave once I modify the data in real time.
Do you think it is possible to use geometric shaders on top of polyvox?
Author:  David Williams [ Tue Feb 24, 2009 1:12 pm ]
Post subject:  Re: 3D surface reconstruction for volumetric voxel data
Gajananan wrote:
Though I have an issue in data size. For 128 ^ 3 data size, I could not get the full object model. What I do now, i have 668-668-469 size data, then cropped it to be 128 ^ 3, this lead to whole in the surface. I am trying to load 256 ^ 3, it leads to memory problem. I have to tackle this problem as Polyvox limit the size to be power of 2.


It's true that PolyVox is pretty memory hungry but you should be able to do 256^3 without a problem. That is the size of the castle volume used in the Thermite demo. I assume that at this point the original 668x668x469 volume is no longer in memory? You only have the PolyVox volume and the resulting mesh in memory?

To be honest, the memory management aspect currently needs a lot of work. Fortuantly I intend to look at it soon (next few weeks) because I need to be able to load large dynamic terrains for the game I'm working on. There are two main areas for improvment:

1) The volume needs to be compressed in memory. It contains large continous regions which should compress down quite well.

2) The size of the meshes needs to be reduced. I think that each vertex is currently about 30 bytes which is too big when there are millions of them.

I will look at both of these as I get time...

Gajananan wrote:
I still want to improve the smoothness. Because I am not sure how it would behave once I modify the data in real time.
Do you think it is possible to use geometric shaders on top of polyvox?


Yes, I believe you can use gemetry shaders. All PolyVox does is give you the mesh data, how you render it is up to you. I have no experience here though, so your on your own! Hopefully there are some tutorials around on using geometry shaders with OpenGL.
Page 3 of 7 All times are UTC
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/