Hi all,

Sorry for the bother!

It seems like currently the git clone --recursive returns some error:

shwang@SJC-PRE-16-5301:/mnt/e$ git clone --recursive https://github.com/hjwdzh/QuadriFlow.git
Cloning into 'QuadriFlow'...
remote: Enumerating objects: 4095, done.
remote: Total 4095 (delta 0), reused 0 (delta 0), pack-reused 4095
Receiving objects: 100% (4095/4095), 16.05 MiB | 15.02 MiB/s, done.
Resolving deltas: 100% (2381/2381), done.
Checking out files: 100% (281/281), done.
Submodule '3rd/libigl' (https://github.com/libigl/libigl) registered for path '3rd/libigl'
Cloning into '/mnt/e/QuadriFlow/3rd/libigl'...
remote: Enumerating objects: 14, done.
remote: Counting objects: 100% (14/14), done.
remote: Compressing objects: 100% (13/13), done.
remote: Total 34988 (delta 1), reused 3 (delta 1), pack-reused 34974
Receiving objects: 100% (34988/34988), 9.17 MiB | 5.73 MiB/s, done.
Resolving deltas: 100% (21287/21287), done.
error: Server does not allow request for unadvertised object 3123b967c1e2cd8e9233937396c5736f2274b5e0
Fetched in submodule path '3rd/libigl', but it did not contain 3123b967c1e2cd8e9233937396c5736f2274b5e0. Direct fetching of that commit failed.

Could you please help take a look when you get a chance?

Thank you!

Best,
Shawn

Great job! But I am interested in anisotropic quad meshes, where can I find the codes for IGM? Does anyone knows? I cannot find that online. Thanks!

hello, i've read your paper: QuadriFlow: A Scalable and Robust Method fo Quadrangulation. but i got puzzled with this two problems:


so how i can get (7) by splitting? i thought (7) should be the 1-norm of difference between the two variables, but why it's the sum?
Another question is that in your source code, you seemed to give up transferring (7) to MCF but select the 1-norm of d as your optimization objective restricting a small change with d-star. so i want to know if i took my wrong way or it's truth.
glad to your answer.

Hi!
Thanks to the work by Pablo Dobarro, we are going to integrate QuadriFlow in Blender with other sculpt mode improvements:
https://www.artstation.com/pablodp606/blog/1vEn/new-blender-sculpt-mode-introduction

Now the reason I've opened this issue is that we (the developers at the Blender Institute) are a bit unsure how we should go about the integration. We are going to need to do some slight modifications to the QuadriFlow code to make it able to interface with the internal mesh formats in blender.

However, we could of course like to have our changes merged upstream too. So we wonder if you are willing to accept application specific APIs like this.

In addition to this, we are uncertain how we should include QuadriFlow in the blender project. We have two options:

1. Copy the Quadriflow code base into the blender code base so that users doesn't have to download and install Quadriflow. (This could lead to the code base in blender and the official code drifting apart as it would effectively be a fork)

2. Make Quadriflow a standalone library and include it as a dependency. (Might be a bit unnecessary if it is only blender that would use the library in question)

Do you guys have any thoughts or input?

Currently, the software has bugs in subdividing meshes with open-boundary, which results in a crash if you turn on the SAT solver.

Better sharp boundary detection algorithm.

I bumped into this app to day and was wondering if there gonna be any release builds for other OS's.
I read on a blender artist thread about BakeMyScan that something was coming. But that was back in November.

Im not familiar with Cbuild or any method of compiling.

I would like to know if an OSX build will be made and if this is version depended on the os status or so.

Perhaps I am missing something, but my manual benchmark suggests Instant Meshes somehow run much faster (5x-10x) than QuadriFlow:

$ time ./instantmeshes -v 2000 -S 2 -d -o test3_output2.obj test3.obj

real	0m2.338s
user	0m4.994s
sys	0m0.198s

$ time ./quadriflow -adaptive -i test3.obj -o test3_output1.obj -f 2000

real	0m23.900s
user	0m22.801s
sys	0m0.883s

$ time ./quadriflow -i test3.obj -o test3_output3.obj -f 2000

real	0m11.733s
user	0m11.051s
sys	0m0.567s

test3.obj is 3bommes/Buddha_input.obj, I have ran a few similar tests, on macOS they are consistently slower than Instant Meshes.

In the center is Instant Meshes' result, Left/Right are QuadriFlow's (Right: -adaptive enabled; Left: -adaptive disabled), all 3 results have tris count at around 15K.

(-mcf and -sat flags increases the compute time further, albeit with better results, so I omit them here.)

Is there an obvious reason as to why this contradicts with the paper's test results?

(The key difference here might be that I am on Intel GPU; but CPU speed is comparable.)

Currently we assume input as a water tight 2-Manifold. We need to additionally deal with open boundaries.

Hi, QuadriFlow is really nice, I am integrating QuadriFlow to my modeling tool.

Is there a way to fetch the source vertex please? What I mean is that,
I get the new generated vertex from O_compact, quads from F_compact. But from where does O_compact generated?
For example, there are color and material infos related with old vertex, how to connect the info with the new vertices?
Thanks!

Hello,

Amazing library! Thanks for sharing.

Could it be possible to get your webGL tool for quadriflow?
Thank you

I'm aware of another similar program called "QuadRemesh" which handles retopology like a beast, but has the caveat of costing 190 dollars or so for a perpetual license. My question is does (or will) this program match that quality? Not to sound entitled or rude or anything lol, but I figured I'd ask.

Hi, I just built a Release 64 Bit Build for Quadriflow for Windows OS. I would like to use it in my JRemesh Blender addon. Is it ok if I upload it to the github in a bin/quadriflow/win64 folder so that the guys can use it? Or do you provide any binaries? Thx and best regards.

I ran ./quadriflow -i ../../icosphere7.obj -o quadsphere.obj -f 131072 on a order 7 icosphere triangular mesh in hopes of returning a 131,072 face quad mesh and it returned a mesh with 1,076,594 face quad mesh. While the quality is fantastic, it's nearly an order of magnitude higher resolution than I'd asked for.

I can understand if there cannot be an exact match to the number of faces due to geometric principals, but this seems way off. Am I missing something about the -f flag?

If it helps, I've attached the input I used as icosphere7.zip and the output as quadsphere.zip.

Some Blender users have managed to find that QuadriFlow produces a very deformed mesh in some cases. The mesh is manifold and does have consistent face normals, so I'm not sure why this is happening:

I've reproduced this issue with the standalone program too. Here is the mesh file and command line arguments to produce this result:

input_mesh.zip
./quadriflow -i bug_quad_remesher.obj -o test.obj -f 11000

It seems like this deformation happens when it actually produces the output mesh geometry as the O matrix contains this mesh after the last step:

So the deformations in the legs doesn't seem to be because of bad field data (?).

Any input on this issue would be greatly appreciated.

Hi,
When I test Quadriflow on ShapeNetCore.V1 dataset, I get following error for many models.

10000 /Volumes/ShapeNetCore.v1/02858304/7bfb0e8e74e1bd2bbe7ca2216dc1576a/model.obj /Volumes/ShapeNetCore.v1/02858304/7bfb0e8e74e1bd2bbe7ca2216dc1576a/model_qflow.obj Initialize... Use 0.024000 seconds Solve Orientation Field... Use 0.011000 seconds Solve for scale... Use 0.001000 seconds Solve for position field... Use 0.022000 seconds Solve index map... wrong init 0 0!

I only set -f -i -o.

Am I forgot to set any other parameter?
Or I am using a wrong dataset. The datasets that I downloaded are ShapeNetCore.V1 and V2.

Thank you.

currently valence can be larger than 5.

Study the reason and fix it.

(Just a question and not a bug report, but GitHub gives no way to send messages to people other than filing a bug. Feel free to close.)

The code and almost all libraries are under permissive licenses, but Eigen/SparseCholesky is LGPL. The build fails with -DEIGEN_MPL2_ONLY. Just wanted to make sure you're aware of it, since it it's a deterrent to using the code and it seems to be the only copyleft dependency.

FYI, it would be helpful to have a clear list of dependencies, since it has a ton of them. Other than the ones listed in the readme, it's looking for at least "ceres", "lemon", "pcg32" and "pss".

This project seems super useful, but it comes without license, meaning nobody can legally use it :) How is this licensed? BSD? MIT? Can you please add a LICENSE or COPYING file?

Not sure what happens...

Error 18 error C2064: term does not evaluate to a function taking 2 arguments d:\test_plugin\quadriflow-master\eigen3\eigen\src\Core\CoreEvaluators.h 347 1 quadriflow

Any chance of sharing release builds for Windows, Linux and Mac?

Dear developers,
Thank you very much for your great software. I have a 2-d triangle mesh with multi-scaled size. I used this mesher to convert to quad mesh, however, the output mesh is uniform mesh with many extra nodes added. Is there any command to let the mesher generate high quality but non-uniform meshes?

Thanks,
Tang Laoya
input.zip

Quadriflow fails to build unless a previously deleted line in arg_parser.cc is reinserted. It's line 224 here: eca2e48#diff-78f17c1e2dfeb45838ed333e9951d1fd

Without that line, the variable s isn't declared in the scope. Is it correct that this line should be reinserted to correctly build?

I am trying to build with CUDA support and get the performance a bit better, but cannot find the "AdjacentMatrix.h" header included in Optimizer.cu on line 3. Can you point me in the right direction? Thanks

It seems like Quadriflow sometimes generates folds in the output mesh:

I've tried most of the toggle-able features, but none seems to eliminate this.
I will attach a mesh that consistently produces this issue:
test_mesh.zip

This is the command I run to generate the output mesh:
./quadriflow -i Bruffles.obj -o test.obj

Is this something that can be fixed or is it an inherent problem with the algorithm?

Hi,
First thanks for this great tool.

I've been trying to make a build of quadriflow for Linux for 4 days now with no success.
I have downloaded the dependencies (Eigen and Boost). But it seems that errors comes from Lemon, which can not find GLPK, ILOG, COIN, SOPLEX. Are thoses librairies pre-requesites ?
Is an issue you run into also ?

Cheers

Yann

Probably solved with an energy minimization after getting the quads.

During dynamic solver, projection can be bad at large curvature.

Hello, Can I use Quadriflow to integrate in Blender? The plugin will be fully opensource Under GPL 2 licence. My objective is just to bridge the QuadriFlow binary with Blender to give it a GUI and also to make it accessible to mass. If it's permitted, please let me know the licence block I need to give in the script and description too. Thanks for this nice algorithm.

Hello,

I have made a simple blender addon which basically invokes the quadriflow.exe (Windows build via MSVC 2017) via commandline. The selected object is converted to obj, processed by quadriflow and return as obj, as well as converted back.
My issue with it is now, that it will exit prematurely in "Build Integer Constraints…" if too many "sharp edges" exist (and if "-sharp" is enabled).

For example, see this input .obj (zipped)
original.zip
Quadriflow.exe is called like
quadriflow.exe -i original.obj -o out.obj -f 500 -sharp

The addon is here:
QuadriflowRemesh.zip

Just install in blender and in the addon preferences, point to a path with the quadriflow executable.
Then...

and enable "Sharp" in the following popup field and "Start Remeshing".
The original blend file :
tower.zip

Meshing the Gargoyle model was very impressive... but why can I not mesh a simple square plate with a hole in 2D? If you can mesh a Gargoyle surely you should be able to mesh this very simple model?

I have attached an example of a 2D quad mesh on this model below to describe what I would like to see. I was hoping QuadriFlow could generate a better mesh on this example.

Triangular Input Mesh (.OBJ File)

My Quad Mesh (hoping QuadriFlow would do something better...)

Certainly with the manifold "work-around" QuadriFlow produces a mesh on an enclosed volume but not a 2D mesh.

QuadriFlow Mesh with Manifold Work-Around (not desired)

Please find the .obj file content pasted below.

# 93 vertex positions
v  0 50 25
v  0 50 0
v  0 34.0483 20.7422
v  0 34.0483 20.7422
v  0 34.8229 23.1262
v  0 50 25
v  0 50 0
v  0 32.7051 18.6258
v  0 34.0483 20.7422
v  0 50 0
v  0 28.6812 15.7022
v  0 30.8779 16.9098
v  0 50 0
v  0 26.2533 15.0789
v  0 28.6812 15.7022
v  0 50 0
v  0 0 0
v  0 23.7467 15.0789
v  0 0 0
v  0 21.3188 15.7022
v  0 23.7467 15.0789
v  0 23.7467 15.0789
v  0 26.2533 15.0789
v  0 50 0
v  0 0 0
v  0 17.2949 18.6258
v  0 19.1221 16.9098
v  0 19.1221 16.9098
v  0 21.3188 15.7022
v  0 0 0
v  0 30.8779 16.9098
v  0 32.7051 18.6258
v  0 50 0
v  0 0 25
v  0 15.1771 23.1262
v  0 15.9517 20.7422
v  0 0 25
v  0 15.0197 25.6279
v  0 15.1771 23.1262
v  0 15.9517 20.7422
v  0 0 0
v  0 0 25
v  0 15.9517 20.7422
v  0 17.2949 18.6258
v  0 0 0
v  0 0 50
v  0 16.5567 30.3583
v  0 15.4894 28.0902
v  0 0 50
v  0 18.1545 32.2897
v  0 16.5567 30.3583
v  0 0 50
v  0 50 50
v  0 25 35
v  0 25 35
v  0 22.5131 34.6858
v  0 0 50
v  0 22.5131 34.6858
v  0 20.1825 33.7631
v  0 0 50
v  0 20.1825 33.7631
v  0 18.1545 32.2897
v  0 0 50
v  0 50 50
v  0 29.8175 33.7631
v  0 27.4869 34.6858
v  0 50 50
v  0 33.4433 30.3583
v  0 31.8455 32.2897
v  0 50 50
v  0 50 25
v  0 34.5106 28.0902
v  0 34.5106 28.0902
v  0 33.4433 30.3583
v  0 50 50
v  0 31.8455 32.2897
v  0 29.8175 33.7631
v  0 50 50
v  0 27.4869 34.6858
v  0 25 35
v  0 50 50
v  0 15.4894 28.0902
v  0 0 25
v  0 0 50
v  0 50 25
v  0 34.9803 25.6279
v  0 34.5106 28.0902
v  0 15.4894 28.0902
v  0 15.0197 25.6279
v  0 0 25
v  0 34.8229 23.1262
v  0 34.9803 25.6279
v  0 50 25

# Mesh with 31 faces
f  1//1 2//2 3//3
f  4//4 5//5 6//6
f  7//7 8//8 9//9
f  10//10 11//11 12//12
f  13//13 14//14 15//15
f  16//16 17//17 18//18
f  19//19 20//20 21//21
f  22//22 23//23 24//24
f  25//25 26//26 27//27
f  28//28 29//29 30//30
f  31//31 32//32 33//33
f  34//34 35//35 36//36
f  37//37 38//38 39//39
f  40//40 41//41 42//42
f  43//43 44//44 45//45
f  46//46 47//47 48//48
f  49//49 50//50 51//51
f  52//52 53//53 54//54
f  55//55 56//56 57//57
f  58//58 59//59 60//60
f  61//61 62//62 63//63
f  64//64 65//65 66//66
f  67//67 68//68 69//69
f  70//70 71//71 72//72
f  73//73 74//74 75//75
f  76//76 77//77 78//78
f  79//79 80//80 81//81
f  82//82 83//83 84//84
f  85//85 86//86 87//87
f  88//88 89//89 90//90
f  91//91 92//92 93//93

Two directions

• Minimum cost flow
• Energy minimization after remeshing.

If there are two orientation singularities at neighboring triangles, it may cause some problems with our current system.

Following are the error messages when I ran cmake -G"Visual Studio 14 2015 Win64" ..:

-- No preference for use of exported Eigen CMake configuration set, and no hints for include directory provided. Defaulting to preferring an installed/exported Eigen CMake configuration if available.
-- Failed to find installed Eigen CMake configuration, searching for Eigen build directories exported with CMake.
-- Found exported Eigen build directory: D:/3rd-parties/eigen-3.3.7/build
CMake Error at cmake/FindEigen.cmake:89 (message):
Failed to find Eigen - Could not find file:
D://include/eigen3/Eigen/src/Core/util/Macros.h containing version
information in Eigen install located at: D://include/eigen3.

The cmake script has successfully found the exported Eigen build directory, but it seems that it then failed to assemble the correct path to Macros.h according to that. The root folder of Eigen is one level above the build directory, i.e., D:/3rd-parties/eigen-3.3.7. As a result, the correct filepath should be D:/3rd-parties/eigen-3.3.7/Eigen/src/Core/util/Macros.h, not D://include/eigen3/Eigen/.... So, could you please fix this issue in FindEigen.cmake? Thanks a lot.

My environment:

• Windows 10
• CMake 3.15.3
• Visual Studio 2015

A good method that produces

1. nonmanifold
2. watertight
3. valence-345
   quad mesh.

The parameter -f [resolution] isn't explained in the readme.

Is resolution the numbero of vertices? Faces? Is it a vertex decimation scale?

You could add a simple example in the readme to explain better what you intend for resolution.

Thank you very much, Riccardo.

Multiresolution Manifold Representation

Hello,
I want to try compil with cmake and msg error: FindEigen

Correctness Measurement

1. Wrong Topology
2. Flip Triangles

Quality Measurement

1. distance from quads to original mesh
2. Edge Variance/longest,smallest ratio
3. Angle between orientation and quad edges
4. Singularity numbers
5. quad angles, etc.

Efficiency Measurement
Maybe make model multi resolution, and give a timing

1. Build Singularity Graph

   *1.1 Thicken the singularity Traceline
   1.2 Connect graph
   1.3 Assign face cell

2. Match graphs

   2.1 Rigid Alignment & Nearest Neighbor
   2.2 Validate Edge (scale)
   2.3 Propagate match

3. Align Geometry

   3.1 Transform the source model According to graph match
   3.2 Refine the alignment

Allow nonuniform edge length for quads.

When the -sharp flag is used, a segmentation fault occurs. The last thing printed before the error occurs is “Solve index map...”, suggesting the error may be occurring in the ComputeIndexMap function.

Can this error be reproduced by other users as well?

Hi,
Thank you for providing this awesome project.
In this project, Adjustment of scale is a very important operation to obtain a beautiful result. However, in the paper, I do not find the detail about the scale adjustment. Hence, I have few questions about it.

1. Is the scale adjustment operation based on a published paper or existing algorithm?
2. In TravelField function, how to understand the following code block.
   MatrixXd &T = triangle_space[f]; VectorXd coord = T * Vector3d(p - V.col(F(0, f))); VectorXd dirs = (T * pt); double lens[3]; lens[0] = -coord.y() / dirs.y(); lens[1] = (1 - coord.x() - coord.y()) / (dirs.x() + dirs.y()); lens[2] = -coord.x() / dirs.x();
3. In EstimateSlope(), could you please teach me why we need area when calculate mK at line 93, and then divide it at line 99.

Thank you for your time,