- From: "Jonathan Bronson" <bronson@sci.utah.edu>
- To: "Petar Petrov" <pip010@gmail.com>
- Cc: <cleaver@sci.utah.edu>, "Moritz Dannhauer" <moritz@sci.utah.edu>
- Subject: Re: [Cleaver] test
- Date: Mon, 30 Jan 2017 10:02:42 -0700
Hi Petar,
This is exciting to hear! My original intent was to maximize
compatibility of Cleaver source code on as many systems as possible.
Since that time, the support for and community efforts around c++11/14
have been tremendous. So I'm all for making use of it and its modern
features.
The high RAM usage in Cleaver is primarily due to the current
implementation
using an in-memory meshing strategy. The input volumes and these
meshes themselves,
can be quite large. And the adjacency information computed and stored
for performing
the core cleaver operations multiples that.
I've prototyped a streaming version of cleaver (in 2d) that only keeps
a small wave front
of the mesh in memory, serializing finalized tetrahedra to disk as
they are ready. It uses
substantially less memory and is also more cache coherent, making it
run faster. I think that's
the right direction to go for running on standard desktops, but I
don't have the bandwidth to
work on that at this time.
I wasn't involved in the BioMesh3D development, so I can't speak too
much on that. Your breakdown
sounds more or less right to me. Perhaps Moritz can be of more
assistance with BioMesh3D?
Feel free to reach out to me at bronson@sci.utah.edu to chat more
about.
Cheers,
Jonathan
On Mon, 30 Jan 2017 13:59:14 +0100
Petar Petrov <pip010@gmail.com> wrote:
Hi Jonathan,
cc Moritz
I am a PhD student who is using the Cleaver2 on both rats&humans
head
models.
Planning to improve upon and share back. For now I am cleaning abit
the
code base and switching to modern c++11/14.
There are plenty of TODO type comments around. Also there seems to
be some
mem-leaks I am investigating now.
The problem is that cleaver2 sucks alot of RAM, making it not
practical to
fetch big volumes. I do manage to produce meshes but only after
running it
on a cluster machine where each node has around 64GB RAM.
On another notice, I am also planning into getting back to
BioMesh3D, which
had the other problem (running too slow) the particle optimization
step was
taking months !!!
So just to summarize: Cleaver2 is RAM bound while BioMesh3D is CPU
bound.
The quality of surfaces in BioMesh3D is better than cleaver2, which
however
has stronger tet-quality guarantees. Also Cleaver2 tend to generate
~
20-30% more elements than BioMesh3D.
Have you been involved in the creation of BioMesh3D too? If not, do
you
know who I can contact.
Anyway, let's try to keep in touch. What is best channel to
communicate?
How about a quick skype anytime soon?
Looking forward to collaborating with you,
cheers Petar
On Thu, May 22, 2014 at 3:21 PM, Jonathan Bronson
<bronson@sci.utah.edu>
wrote:
Success! :)
On Thu, 22 May 2014 14:15:37 +0200
Petar Petrov <pip010@gmail.com> wrote:
First try to write to the public mailing list.
Not sure If subscription went all fine.
Please someone reply! :)
--
All the best,
Petar Petrov
http://ppetrov.net
- Re: [Cleaver] test, Petar Petrov, 01/30/2017
- Re: [Cleaver] test, Jonathan Bronson, 01/30/2017
Archive powered by MHonArc 2.6.18.