Local Curvature Entropy for Saliency-Weighted Simplification
This is a small project page for my Eurographics 2016 short paper "Mesh Saliency via Local Curvature Entropy" [LKF16]*. I finally found time to build a 3D Web demo, giving the impression of a tool where an artist can balance between pure quadric-based mesh simplification (as in the seminal paper of Garland and Heckbert [GH97]) and our saliency-weighted simplification, using saliency scores computed by our method.
The demo shows, for each of the four meshes, a split-screen view that compares the original mesh to a simplified version, consisting of about 1,000 triangles. In addition, you can play with the overall weighting of the saliency values against the quadric errors. You will notice that the algorithm works well for some meshes and less well for others. In particular, for the Bimba mesh, it wastes a lot of vertices for the knot in the hair, since this structure has a complex shape and is therefore classified as salient. On the other hand, for the woman and wolf meshes, it helps to improve the visual results over the classic method (if the weighting is not too strong).
While this is not a live simplification application, but just a viewer that shows a series of precomputed results, it should be possible to implement a similar technique inside interactive 3D modeling applications. This is especially possible because saliency estimation, using our method, and weighted mesh simplification can already be performed at interactive rates, for a lot of small and medium-sized meshes. Click the following image link to check out the demo:
Source Code (MIT License)
If you want to use the algorithm in your application, or if you want to study the source code for research purposes, you are welcome to download it here:DOWNLOAD SOURCE CODE (ZIP ARCHIVE)
If you would like to use the code under another license, just let me know. Also, if you include the method (or a modified version) in any of your software, research work, frameworks, I will be happy if you tell me via email (max.limper( at )siggraph.org).
Finally, if you want to play around with the results, the supplemental material of the paper might be useful, as it contains meshes in PLY format which have per-vertex saliency values (in normalized range [0, 1]).DOWNLOAD SUPPLEMENTAL MATERIAL (ZIP ARCHIVE)
- [GH97] M. Garland, P. Heckbert: Surface Simplification Using Quadric Error Metrics. Proc. 24th Annual Conference on Computer Graphics and Interactive Techniques (SIGGRAPH 1997)
- [LKF16] M. Limper, A. Kuijper, D. W. Fellner: Mesh Saliency via Local Curvature Entropy. Proc. 36th Annual Conference of the European Association for Computer Graphics. (Eurographics 2016)