|
From: Michele <eh...@gm...> - 2008-03-26 22:55:11
|
> About the curvature: Fabio has committed a new curvature component for > the vertices (they are inside the vertex plus components). > It will part of meshlab1.2. Until the release of the 1.1.1 (matter of > days i am waiting for a confirmation of the last bug removals) > concentrate on the planar flip polygons. In the mean time Implement also > the Laplacian smooth that do not moves vertices far from the surface > (e.g. laplacian smooth but only vertices that fall withing a given > epsilon distance from all the original faces are accpeted, so vertices > moves only in planar regions and along creases. Ok, now I will concentrate on the planar flips and laplacian smooth which move vertices less than a certain threshold (or zero). For planar flips, I would like to introduce an additional parameter to PlanarEdgeFlip template in order to introduce planar flips who use different metrics for the quality of triangles (using vcg: Quality, vcg: QualityRadii and vcg: QualityMeanRatio, as in plugin meshcolorize). These days I read the new code for vertex curvature (/vcg/complex/trimesh/update/curvature.h, and vcg/simplex/vertexplus/component.h), and I have also given a look at the paper quoted the source (Discrete-Differential Geometry Operators for Triangulated 2-Manifolds). The fact is that the existing code calculates the Gaussian curvature and the Mean curvature normalized, while the paper on the flip based on curvature (http://www2.in.tu-clausthal.de/~hormann/papers/Dyn.2001. OTU.pdf), the curvature metrics are calculated starting from K and H not standardised. In a previous mail I mentioned the possibility of avoiding the use of VF adjacency for the implementation of mesh optimization by edge flips based on local curvature. To do this without the VF adjacency should: 1) Initiate the values of curvature (K, H, A) on vertices (FF adjacency only). 2) During the flip, update values (K, H, A) considering only the faces that change with the flip. This approach is only possible if you know K and H (not normalized) and A. With H and K normalized and A unknown, it is always necessary to recalculate the curvature on vertex (walking around the vertex with VF adjacency), whenever the curvature on vertex change. So I wanted to ask if you can give K and H not normalized, and A, as components. With these items you can get both versions normalized, and the three metrics described in the paper curvature of the link. > About the combination of the various flipping strategy i am not so sure. > I would like to make some experiment before. > Lets concentrate on a robust implementation of the single strategies. I > see that strange things can happens sometimes when applying multiple > times the same strategy... I recently made a change in vcg/complex/local_optimization/tri_edge_flip.h that avoids degenerating flip (which generate two overlapping faces identical). This could be the situation that created the "strange things" after several run. The change has been applied to both the planar flips and the flip based on curvature. Michele Onnis |
