FYI, the “full announcement” link is broken (missing “http://”).

1. If using without gradient, HMPC generates a “blocky” surfaces. However, using the simplest central 5-point residual derivatives applyed to a given scalar field to generate a gradient solves this problem.

2. If using HPMC without a gradient, an algorithm generates a lot of zero-area triangles (can be 5 times more than “real” triangles). Indeed, this should fixed.

3. HPMC does not generate indexed triangles, which are needed to export data from HMPC program to MatLab, 3DMax etc., also to separate isosurfaces. The straight algorithm for indexing a million of vertices leads to 1e12 operations and takes hours. To resolve this problem I had to create a qsort based algorithm, which grows as Nlog(N) (not N*N) and can solve this problem in seconds on CPU. Nevertheless, an indexation of triangles is an actual problem of the HPMC library and a real-time GPU imdexing is strongly needed.

4. Another problem of HPMC isosurfaces, which complicates its practical application is a really huge number of triangles, which is proportional to volumetric mesh size. Therefore a triangles reduce algorithm should be included into HPMC library. For the moment I have only an idea of this algorithm: index triangles -> separate isosurfaces -> generate a bezier surface -> apply tesselation.

5. The unsigned byte type for each grid point in the texture can be changed within a texture definition without any hack of the HPMC code. Each of the RGBA components are casted to float incide of HPMC, so int textures is the physical limit of HPMC. (There is no word about this in documentation).

Finally. The HPMC library is very simple, and fast. Having a simple and self explaning examples. I do recommend this lib for everyone who needs a real time calculation of isosurfaces, especially if there is no need in its subsequent analysis.

I’m opened to any discussions at: reaper@mail.univ.kiev.ua