Jump to content

  • Log In with Google      Sign In   
  • Create Account

Fixing terrain tearing


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
4 replies to this topic

#1 PAndersson   Members   -  Reputation: 394

Like
0Likes
Like

Posted 07 October 2012 - 04:56 AM

What you see in the attached screenshot, besides an early test of my GUI system, is a an equadistance point sphere with a heightmapped applied to it. Tearing artifacts are highly visible (at least in motion :P) and I would like some suggestions in how I would go about eliminating them.

The sphere is essentially a cube with its vertices remapped into a spherical shape, and the cube is constructed from 6 quads each consisting of n*n points. These quads are then attached together into a single mesh before the remapping. The tearing appears at the 'seams' of the six quads, as they do not actually share any vertices. Having the quads share vertices solves the tearing issue, but introduces problems when it comes to texture coordinates. I was thinking about detecting overlapping vertices and creating quads to patch these up. While detecting overlapping vertices are easy enough, I'm unsure about the best way to actually create indices for them without teh result being massive and overcomplicated code.

tearing.png

Sponsor:

#2 Krohm   Crossbones+   -  Reputation: 3117

Like
0Likes
Like

Posted 08 October 2012 - 01:08 AM

an equadistance point sphere with a heightmapped applied to it ... The sphere is essentially a cube with its vertices remapped into a spherical shape, and the cube is constructed from 6 quads each consisting of n*n points.

So it's a cube. It would have been nice to show it without displacement at least. But I guess it's a cube.
What you have is the obvious result of stitching 6 independent heightfields near each other. This is not a typical terrain system and I have no idea why you're doing that but anyway... solution, solution, solution.

If the heightmaps are static then they're effectively a decoupled representation of the vertices themselves for reduced storage. You know in advance where each vertex will end. Just create some "skirts" using cube edges from different faces and you'll be set.
If the heightmaps are dynamic... I would switch to doing things right because you're doing for no apparent reason IMHO.

#3 PAndersson   Members   -  Reputation: 394

Like
0Likes
Like

Posted 08 October 2012 - 01:17 PM

To clarify, it is essentially six quads arranged as a cube and then changed into a sphere. At 12 x 12 vertices per quad, the mesh looks perfectly spherical without any displacement or heightmaps added.

Using spheres for the terrain is the best option, as the game I'm working on is a 4x space strategy game with quite a bit of ground action and I have a hard time seeing any better way to represent planets then to start with spheres. This way, it will easy to represent the planets at multiple levels of zooming and from any angle. Of course, the displacement shown in my screenshot is way beyond what you will find in the game, but the less extreme it is the harder the tearing is to see and it is impossible without any displacement at all.

Edited by PAndersson, 08 October 2012 - 01:19 PM.


#4 Krohm   Crossbones+   -  Reputation: 3117

Like
1Likes
Like

Posted 10 October 2012 - 12:47 AM

If the faces are disjoint, they are likely to cause this problem (much more if the normal isn't the same).
If the heightmaps are disjoint, they are likely to cause this problem.
If the independent patches have independant LOD they will be worse.

Either add skirts or make your mesh continuous (it does not really have to share vertices, but it must at least use numerically equal vertices, end even this is not guaranteed to produce the same rasterization).

Solidifying independent meshes is a serious task.

#5 PAndersson   Members   -  Reputation: 394

Like
0Likes
Like

Posted 11 October 2012 - 11:43 AM

Solidifying independent meshes is a serious task.


I have noticed. Anyway, they are disjoint but can easily be made not disjoint as I mentioned before (detecting overlapping vertices is easy enough) but that introduces texture coordiante issues.

Each sphere has only one heightmap, and really needs nothing else. Each sphere is treated as a single object in regards to LOD (though none of them will be particulary demanding when it comes to rendering).

I managed to eliminate most issues by picking another UV-mapping strategy. Though it still indtroduces a single vertical seam between the vertices that have an U-coordinate of around 1.0 and 0.0, as the fragments betwee these end up interpolating that coordinate between those two extreme values. This does not matter for the heightmap (as they are applied per vertice and no interpolated texture sampling is done) but is very visible for other textures. I was thinking of running my UV-mapping formula in the fragment shader and thus calculating texture-coordinates per fragment instead. This would eliminate the problem, but as the formula contains somewhat expensive trigonometric functions I'm afriad it can have a performance hit.


EDIT: A quick test seems to confirm that the seam vanishes (provided I use a texture that tiles!) with no noticible performance hit so I might go with this method...

Edited by PAndersson, 11 October 2012 - 12:21 PM.





Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS