CGAL users discussion list

[Jean-Sébastien Franco <>]

On 31 Oct 2006, at 3:01, Wonwoo Lee wrote:

> Thank for your reply Franco.
> I know your EPVH library, but I have not tested
> because I'm working on Windows platform.
> I read CGAL document about nef polyhedron.
> It provides what I wanted to do.
> Does visual hull reconstruction take much time with many views ?
> Did you measure the computation time to reconstruct
> visual hull with CGAL and with EPVH ?

I didn't do rigorous sets of measurements per se.
But (correct me if I'm wrong) the time for a few views with something  
like 100 vertices per view silhouette is in the order of minutes in  
CGAL.

I had a 4-view example of a person (200 vertices per silhouette) that  
I had tested and it took about 6-7 minutes on my machine at the time  
(2.6 GHz I believe), provided I didn't have LEDA. After discussions  
with Peter I think it came out that LEDA would probably improve the  
speed by a factor of 2 or 3 compared to the speediest free arithmetic  
kernel I was able to use, the homogeneous kernel with Gmpz(?) if I  
remember right.

For that same example EPVH produces a manifold polyhedron in ~100ms.  
Note of course that it is an unfair comparison because CGAL and EPVH  
are very different tools and CGAL's nef_polyhedron covers a much  
broader functionality while EPVH is completely dedicated to  
polyhedral visual hulls for a dozen to a few hundred views. I also  
didn't try a broader range of measurements.

On 31 Oct 2006, at 5:05, Peter Hachenberger wrote:
> Hi,
>
> the newest CGAL version features a demo program for the visual hull
> computation (visual_hull.C). You can find it in the directory
> demo/Nef_3/ and is also briefly documented in the README file. Try it
> with the input file mpi.vsh. This file defines the cones of three
> letters and the demo programm intersects them and displays the result.
>
> Peter

Very interesting! I have to test this.

Thanks for the info,
JS Franco