Full Version: Auto-retopology algorithms compared

Show messages:  1-16  17-36  37-56  57-65

Hi Michael,

OK. I thought it could be an ideal algorithm to implement with regard to your plans for polygon to NURBS conversion inside MoI.

I was thinking of an import dialog offering to retopologize an imported polygon mesh using the algorithm, and add the option of subdivision like in the tools by Max, comparable to Stefan's bunny test.

On the other side, the algorithm could offer perfect quad-poly topology output inside the mesh export dialog, ready for subdivision in a polygon-based 3D editor.

---

Hi Metin, from a quick glance it looks like it's more oriented towards processing organic shapes. I'm not sure what would happen if you tried to feed it a CAD model with a mixture of large and small features in it.

- Michael

---

Good point, Michael.

Looking at the Siggraph paper...

http://stanford.edu/%7Ejingweih/papers/quadriflow/quadriflow.pdf

... the algorithm includes a sharp edge detection feature:

Image Attachments:
Screen Shot 2018-08-29 at 12.48.12.png 

---

Hi Metin,

I understand. ;-) Maybe this topic should be posted at the 3D Coat forum, because
i think they have also Instant Meshes now included in the latest beta(s) of 3D Coat.

Maybe Andrew is faster than the Blender folks?!

Regards
Stefan

---

Good suggestion, Stefan, thanks. I'm going to post it at the 3D-Coat forum as well.

---

Hi Metin,

> ... the algorithm includes a sharp edge detection feature:

Yes that probably helps some CAD models from just being completely destroyed but I'd think that having a uniform size output is not going to work well with many mechanical objects.

It would be good to see the result on something like this for example:
https://grabcad.com/library/engine-v-twin-4-valve-heads-1

In the paper there is one screenshot of a motorcycle that was converted and it looks pretty weird to me, many areas are mutated pretty significantly.

This isn't to say at all that it's not a useful tool when used where it is intended.

- Michael

---

Hi Metin, I've attached an example CAD model here (taken from https://grabcad.com/library/ngoc-mai-tug-boat-1), it would be interesting to see what the output is on these:



- Michael

Attachments:
QuadTest_3dm.zip

Image Attachments:
quad_test_screenshot.jpg 

---

Hi Michael,

I've ran your example through QuadriFlow using these parameters:

quadriflow -i examples/Untitled.obj -o examples/untitled_quadriflow.obj -f 3200

The result is pretty bad, and that's an understatement (first one comes from N-gon OBJ, the second from Tri OBJ):





Adding the -sharp option gave me a segfault on the first, and abort trap error on the second run. So no results at all.

Too bad ...

- Fabien

Image Attachments:
Schermafbeelding 2018-08-30 om 09.28.33.png  Schermafbeelding 2018-08-30 om 09.38.30.png 

---

Hi Fabien, thanks for testing it. That definitely shows to me that it would be really weird to try and bundle that directly with MoI, since that's the kind of stuff that MoI is good at making.

And that doesn't mean that QuadriFlow is bad, just that it is not intended to work on CAD models.

It's also really cool that the authors set it up to compile on multiple platforms and released it under a permissive license as well.

- Michael

---

Come on guys, this algorithm turns technical models into art, what more could you ask for? ;)

OK, I'm convinced. Please keep your polygon mesher as it is, Michael! :D

It's a pity though. For a moment I thought Quadriflow could form the ideal connection between NURBS and quad-poly subdivision modeling.

---

Hi Metin, yes over the years I have learned that you often need to take research paper results with a grain of salt, you can't just assume that the results that are shown apply to every situation. Which is fine though, it shouldn't be required that only things that give perfect results in all cases should be published.

But adopting it as a built in feature that users will get surprised by when they use it is a pretty different matter.

- Michael

---

Agreed!

---

At least it works good with organic shapes and maybe they improve it for hard surface cases.

Regards
Stefan

---

Hi Stefan,

Quadriflow is definitely worth keeping an eye on.

Did you get the sharp edge option to work? I'm curious how different hard-surface models are retopologized with that option.

---

Hi Metin,

i exported from MoI as high-density triangle mesh and was able to
use QuadrFlow with the -sharp option on the first model. But in
QuadriFlow i used also a higher poly count setting. The second
model failed here too. But for me it makes not much sense to
use also a high poly count setting, to only have many small quads.

But this is no problem for me, because i have ZBrush, in case
i need to auto-retopo a CAD mesh... :-)

Regards
Stefan

---

I'm not aware of any auto-retopo process for CAD meshes in ZBrush, can you elaborate? I was under the impression using poly groups was required for good results, which means a lot of manual work to break up the surfaces.

---

Hi Chris,

Paul Gaboury showed once in a video, how to poly group a CAD model fast.

https://www.youtube.com/watch?time_continue=1&v=IzFZfJDQOOM

starts at 39:04.

Then you have also the option with MoI to create with a script random
style colors to a model and can use these for poly groups in ZBrush.

But to be honest why to auto-retopo a *MoI* CAD mesh... MoI has the
best mesher available! :-)

Regards
Stefan

---

Thanks for that link. I'll have to watch it several more times.

---

Hi Chris,

Here's an image by a company called Kermaco I once found on Pinterest, explaining a multi-stage CAD model retopology process in ZBrush:



I expect you'll have to open the image in a new tab to see the full-size version.

Image Attachments:
32f4c6a5c66848e3a5f4c49a4abf780f.jpg 

---

Hi Metin and alll,

have you already tried ZRemesher v3? I have not downloaded yet, but i am
very curious how it would remesh the models Michael has posted.

Regards
Stefan

---

Show messages:  1-16  17-36  37-56  57-65