Quad Remeshing in Blender - GSoC 2014 Suggestions

As some of you may know one of the accepted projects for the 2014 Blender GSoC is a quad remeshing tool. The author has asked for suggestions leading up to the start of the project from users. I’m starting this thread as a place where we can discuss exiting retopology methods, applicable academic papers, and ideas for improvement. I’d ask for people familiar with existing retopo tools (and their pitfalls) to chime in with their thoughts. This is NOT a thread to bicker about who has the best solution/play favorites with software. This is a thread to discuss the development of a new and useful tool within Blender. Any fighting will be immediately deleted.

To get the ball rolling, I’ve created a design document that goes over some of the existing tools, their issues, and the features that I believe would push any remeshing tool into the realm of “viable for animation”. Thoughts and comments are welcome.

Two excellent papers on mesh generation.

I think the remesher paper which is going to be included/worked on for GSOC actually gives good enough results with no spirals?

if you could paint those black cylindrical lines yourself, you would be set with a powerful tool…

I like your idea with the Grid suggestion tool!

anohter feature I would love to see in a remesher tool would be
if it could use the manual retopo form the artist and would continue that or use that as a starting point.
a way that the remesher ignores areas which are marked as no remesh would also help.

The results in that paper are pretty abysmal using the post process indicated. Before post process it results in countless ngons, and the solution in the paper is just to triangulate everything. Both meshes in that image show some pretty bad twisting (camel’s neck) and ngon creation.

oh you are right, when you look upclose the meshes they are indeed having ngons. I thought the post process was done automatically from the tool.

with Zremesher you can also use Polygroups and curves and the useage of the “Dam_standard” Brush on top of the sculpt also helps to get more predictable results.

converted to Blender that would mean;
adding a nice way with the knife tool to add polygroups which are forcing the algorithm to use those polygroups as a guide where at the end of the polygroups must be an edgeflow.

Matt summed up my concerns nicely in his PDF, at least regarding organic retopo.

The only other suggestion I’d add is manual placement of 3-, 4- and 5-valent vertices:
(only 3 and 5 shown here)

This would facilitate retopo for more complex hard surfaces.

These two Papers are GREAT! I’m pretty sure that Blender users will be happy, to see any of those be implemented.

Agreed. Let’s hope that Alexander is listening us. :slight_smile:

Great PDF m9105826 :wink:

On zbrush with polygroups and the slicecurv we can have more control, but its difficult to have a really good wire, we need to make lot of tests and cut the model in polygroups to do better work.

I make retopo every days with 3D scans on blender and for me an automatic retopo needs to be drived, with curves, painting (weigth paint ?), polygroups or someting else.
There are a lot of possibilities and I hope the dev will listen us to make a great and usefull tool ^^

Don’t forget this one !

Yep, that patch-based editing looks like it’d pretty much address all the requirements. It’s much quicker to use than manual retopo and gives excellent control over the final product.

From that 2013 report it seems that the sowftware is ready, so why is it not published to the wide audience? Is it just for Disney?
About Blender’s auto retopo: let’s just have something first, but later I’d like to see focus on making hardsurface retopo possible. ( a way to mark hardedges on high poly which would then be reserved by the low poly.) Zremesh is quite useless for this kind of job.

well the proposal has already been made. The gsoc student has already researched the paper.

Can we really change the course of development (which paper is used) at this point?

What were the reasons that particular paper was chosen by the student?

Are there any reasons the patch approach is not used instead?

any remeshing tool would be welcome for me! The one on his paper seems quite useful for sculpting!!!
It seems faster to set up than the patch approach. Better for sculpting, but not as good (controlled) as patch for actual retopology.

So what we are really doing here is asking for a slightly different purpose tool than the proposed in the paper.
Wasn’t Jonathan already working on better retopology tools? :smiley:

Please appreciate the spiral remesh paper guys. It’s million years better than what we currently have in b3d for remeshing!

The methods and algorithms for the patch-based method are posted across two papers. Seems like they’re holding the actual implementation close to their chests though, likely because they realize how much money it can make them.

blurymind, those are all good questions. Things are picked up and dropped over the course of a GSoC all of the time. Thomas’ has several “maybe” goals lined up to code if anything he’s researching turns out the be a dead end, for example.

Why that paper was chosen is a good question. It’s ancient in terms of CG papers (2004) and has been long displaced by methods that are faster and more robust without relying on a triangulation post process to cover shortcomings of the algorithm itself. I guess we’ll have to wait for the student to weigh in himself on this one.

the n-gons would be a problem!!

This is an interesting thread. I am the mentor for this GSoC project and am reading it carefully. Thanks for the PDF, m9105826, and thanks everyone else for the links and suggestions.

Ok, so, can you give us some informations, the student will make exactly like the paper or he will be open to discution ?

Regarding the Zimmer-Bommes papers, one thing that may or may not be useful is that they open sourced their constrained mixed integer solver under gpl:


The library only depends on GMM++ and Eigen3. It’s not actually the whole quadrangulation implementation, but as I understand it, it’s an important part.

Hi all, i am Alexander the GSOC2014 developer

I am open to discution always.

I was reading there and many other papers. but i choose the paper of Dong et. al., because i can do this work in the Gsoc2014.
Some other papers are awesome, but need experience in that algorithms, and more people writing code.

With the actual paper the scalar field generated by the user inputs, is really customizable.

I do not want to rush and change the paper without having done some tests and see the results in Blender.
I hasten to perform some tests in Blender, to decide whether to continue with the paper or should change.
I made some test in Matlab, and the results are very predictable.

Thank you all for your opinions and questions.
I try to take into account all opinions and suggestions.

for me the most beautiful thing would be some manual vertices that command, and the rest of the topology is “interpolated” between them. This way you could sort of “pin” vertices like in the uv editor, and move them around until the resulting topology satisfies you.
Something like that, interactivity is a beautiful thing, but maybe a little too imposible with geometry generation?? in the end it could be like the grid fill but as a modifier (hey that’s not such a bad idea…)

i will donate some of my sculpt for testing this is the best way to see the result by experiment whit different sculpt topology and hard surface


I too have quite a bit of re-topo under my belt, going back since 2007.

That ‘Sketch-Based Generation and Editing of Quad Meshes’ is the best go at automatic re-topo I had come across. That’s the way to go in my opinion.