Polygon smoothing - Linking tesselation to Facing ratio?

Matt B

Newcomer
Hi guys, i just had a bit of a random thought earlier to do with improving efficiency in polygon smooth/tesselation operations. I know that it's largely an unused technology and has a whole host of problems but I was wondering if it would be possible/feasible to have an adaptive tesselation technique on a model which is linked up to the facing ratio of each polygon (direction of normal relative to the camera). The further the normal is pointing away from the camera the higher the tessellation on that polygon so that effectively you are largely only doing extra calculations near the profiles where it is actually needed. Of course you have to do extra checking calculations as well as retessellating the model every frame...

I'm sure this has been thought of before and wouldn't work for a number of reasons, probably both unfeasible and too slow. Anyone got any thoughts
 
Since no one else has responded.

Yes it's possible and yes it has been done, I believe Hoppes has a paper using progressive meshes that does something like this.

The problem like pretty much all dynamic tesselation algorythms is that it's often just faster to draw the more complex model than it is to dynamically create a lower poly count one.

More often than not geometry throughput isn't a significant bottleneck.
 
I guess this sort of thing would be more interesting to explore if/when some sort of programmable primitive processor finally makes it onto the GPU.
 
MfA said:
Would be more interesting in the context of displacement mapping as in Michael Dogget's paper.
Thanks for the link. I obviously haven't had time to look over it in detail but, presumably, figure 6 suggests that it won't behave nicely if you need to gradually change the LOD?
 
How do you mean? The LOD is implicit in the parameters, you wont be setting LOD yourself. Also it wont form cracks if that is what you mean, shared edges share splits.

Reverend, you are wrong though ... this wasnt actually the right paper, this discussion reminded me of this paper. It is along the same lines, but a little more recent.
 
MfA said:
How do you mean? The LOD is implicit in the parameters, you wont be setting LOD yourself. Also it wont form cracks if that is what you mean, shared edges share splits.
I thought I read that they take the image size of the object into account and so if you zoom in/out you will probably get a change in the level of tessellation.
 
Back
Top