New Free Automatic and Semi-Automatic Retopology Program

[TimmyZDesign]

I just found out about this free program written by researchers at the Swiss Federal Institute of Technology in Zurich. It performs automatic retopology and also guided semi-automatic retopology. It was demonstrated at SIGGRAPH Asia 2015. You can download it from GitHub for free (the actual compiled executable app is there). The license there says that you can use it for any purpose (including commercial purposes), also see the license there for any further questions. I haven't had a chance to try it yet, but I am interested to see how it compares to 3D-Coat's Autopo or Zbrush's Zremesher.

 

Demo video:

https://www.youtube.com/watch?v=U6wtw6W4x3I

 

Download from GitHub here:

https://github.com/wjakob/instant-meshes

 

I will try to test it sometime this week if/when I get time. If you test it out, please post your results in this thread!

[L'Ancien Regime]

http://igl.ethz.ch/projects/sketch-retopo/

 

It's wicked fast TimmyZ...faster than Zbrush Remesher but I'd have to do some experiments on the mesh qualitieis.

Edited November 18, 2015 by L'Ancien Regime

[Rebelismo]

Thanks for posting! I tried the auto function on some dense 3d scans, but the results weren't too promising. I was able to crash the program multiple times, and Maya couldn't read some of the heavier .obj files. I'll try it a few more times on lighter meshes to see if I can use it in my workflow, as a helper tool to ZRemesher and Autopo. 

[TimmyZDesign]

It looks like this new autoretopology program has attracted a lot of attention!

 

There is a thread about it on Blender Artists (including tests and comparisons of it with Zremesher):

 

http://blenderartists.org/forum/showthread.php?385344-Instant-field-aligned-meshes

 

And this thread about it on the Modo forum:

 

http://community.thefoundry.co.uk/discussion/topic.aspx?f=4&t=117406

 

And these articles on CG Channel and CG Press:

 

http://cgpress.org/archives/instant-field-aligned-meshes-paper-and-standalone-tool.html

 

http://www.cgchannel.com/2015/11/download-free-auto-retopology-tool-instant-meshes/

 

There are also users asking for it to be integrated into 3D-Coat here on this forum:

 

http://3d-coat.com/forum/index.php?showtopic=18580

 

http://3d-coat.com/forum/index.php?showtopic=7912&p=128916

 

It seems to me that everybody is interested in Instant Mesh because it allows the user to have more control of how automatic retopology is created. Currently both 3D-Coat and Zbrush force the user to run the automatic algorithm and wait for results, and then if the results aren't satisfactory, the user alters settings and guides, then tries it again, hoping for better results. The difference with Instant Mesh is that the user can tweak the results as they are generated, allowing for more manual control, while still keeping the time-consuming parts of the retopology process automatic. I really hope Andrew and Raul have a look at this, because this is the kind of semi-automatic retopology I have been wanting in 3D-Coat for a long time, and I think this is the future of 3d-modeling!

I mean, seriously think about it, a big part of what makes traditional subdivision surface modeling difficult is the manipulation of vertices edges and faces. Just selecting vertices, moving them around, and welding them to other vertices is very tedious and time-consuming. 3D-Coat is great because it allows artists to stop worrying about vertices for most of the creation process, until the very end, when retopology is done. But wouldn't it be nice if we still didn't have to worry about vertices even for retopology? I think this is the missing link. If we could get a semi-automatic retopology feature that was so good that it gave us the exact edge flow we wanted while doing most of the heavy work for us, then maybe we could finally stop pushing and pulling verts altogether! I realize that getting a perfect result from automatic retopology is currently very difficult, but it seems to me that if the user can control it and guide it as it happens, then that would be the fastest and easiest way to get desired results.

In any case, I'm glad Instant Mesh was created because it will hopefully inspire the right kind of innovation that we need for the retopology process.

[Tarby]

The only problem I've had is with it hanging when exporting a mesh that has more than one poly island, plus the occasional hang on export with a single poly island mesh. Other than that, though, it's absolutely fantastic and intuitive and definitely the way forward. For a smooth mesh (the ones I've been testing have lots of hard surfaces, protrusions and nooks and crannies) it's an absolute breeze. Top notch.

[Aleksey]

i dunno, it looks awesome in videos. But i struggle to to actually control it. the yellow dots are mostly stuck. and i cant get a decent face mesh out of it.

 

also it doesnt support symmetry.

 

but the live preview of the edge flows is awesome. And being able to define hard edges like that is also great. maybe with more practice i'll be able to get better results.

[L'Ancien Regime]

i dunno, it looks awesome in videos. But i struggle to to actually control it. the yellow dots are mostly stuck. and i cant get a decent face mesh out of it.

 

also it doesnt support symmetry.

 

but the live preview of the edge flows is awesome. And being able to define hard edges like that is also great. maybe with more practice i'll be able to get better results.

 

 

I'd like to see if you can set the density of the mesh you're creating too. Right now it creates a fairly dense mesh.

[Tarby]

There's a slider for mesh density, but at this stage you need to reload the obj when it's changed. You can output an even higher rez quad mesh by switching on 'use dense quads' in the mesh output dialogue.

[L'Ancien Regime]

it's pretty good actually... damn fine

 

 

 

If this was open source I'd be happy to see Andrew write this solution into 3d Coat

 

 

There is a bit of spiraling but not much.

 

 

 

 

On closer examination is does have rougher edges than Zremesher. Zremesher is more fluidly contoured to the object's surface than Instant Meshes. I think you'd have to use more polys to get a usable mesh with Instand meshes.

 

 

Edited November 22, 2015 by L'Ancien Regime

[L'Ancien Regime]

Here is the 3D Coat result, set to 6500 polys and taking 55 minutes 

 

 

 

 

And I publish this with all due respects to Andrew; he was the first to come up with autoretopo working alone, and even that Swiss team of academics couldn't measure up to Zremesher and Pixologic's substantial investment.

 

Also a quick run over the autopo mesh shows  no evidence of spiraling with the edge loop tool for UV mapping.

Edited November 22, 2015 by L'Ancien Regime

[digman]

Instant Meshes... Does a great job on following the edges of hard surface objects... The down side of instant meshes is it creates a lot on N-gons plus has a lot of termination areas disrupting edge loops. It creates all even polygons through out the model regardless if it needs them or not in some areas.

 

I really do like the job it does on hard surface models as it keeps the edges very well so far in my limited testing. If they keep working on the routine after awhile they will have a bang up auto-routine.

 

I wish Andrew would set aside more development time as he is so close himself to having a very good auto- retopo routine... He spent about a week on it a few months ago and improved it but still it is a hit and miss routine at times. I know how to get a clean model to run the routine on, so my problems are not from having dirty meshes. The routine at times still adds more polygons where they are not needed and less where they are needed. Polygons still will overlap each other in areas and painting density still on my end, just really never works... I tried a bunch of times... Maybe sometimes it works. I just tend not to use it now. If Andrew can solve the last problems with auto-retopo plus really make the routine work with hard surfaces then well, that would be just fantastic.

 

Ran the ship through both routines with no setting changes and no guide stokes..

In the picture::

Left side --- Instant meshes--- about a minute in time from start to finish.  5,458 polygons of those 279 were triangles and 59 n-gons

Right side--- 3DC--- about 2 minutes from start to finish. 5018 polygons of those 4 were triangles (inside the gun barrels) and 0 n-gons.This time 3DC did a good job plus the general edge flow was better...

Edited November 22, 2015 by digman

[Carlosan]

3DC Autoretopology is very "settings sensitive" for every model.

 

User need to understand AutopoParams to get a good result, to obtain less spiral and irregularities, but numeric values ranges influence and how every parameter interact with the others will be more clear-easy.

 

 

Instant Meshes on the other side, easy pre-result visualization, give fast result and have a good predictive mesh_flow algorithm. But that speed have a down side -Digman is right-, edge interrupted and irregularities.

[Tarby]

I've managed to cleanup edges in InstantMeshes by going back and stroking the edges after examining the mesh extract. No need to start from scratch and almost immediate feedack. The occasional tris don't bother me, either, but perhaps some workflows require pure quads. I have tools in my main app (Softimage) that can quickly identify and quadrangulate or triangulate the offending culprits. This tech inside 3DC would be awesome. I suspect it'll be appearing in a bunch of apps.

 

Nice comparisons, I have to say.

[TimmyZDesign]

I've managed to cleanup edges in InstantMeshes by going back and stroking the edges after examining the mesh extract. No need to start from scratch and almost immediate feedack.

Yes, this is exactly why Instant Mesh is better than both 3DC and Zbrush auto retopo.

Instant Mesh provides good initial results, and then you take a little time to fix up the bad spots by adding some stroke guides (simply brush on the mesh and it updates), and/or move the yellow "singularity" points to get better edgeflow (also with instant feedback). This ability to edit the mesh with instant feedback is the key point that both 3DC and Zbrush lack. It's much better to quickly fix up the mesh after you see the initial results, instead starting all over again, changing settings, then waiting and hoping for better results.

[Carlosan]

This ability to edit the mesh with instant feedback is the key point.

+1

[Tarby]

+2

[kenmo]

I've been playing with InstantMeshes and for speed alone InstantMeshes blows the doors off of 3DC. It would be like a drag race between nitro burning GMC Supercharged 426 Hemi dragster and a tricycle.

Autopo has a long way to go to match InstantMeshes at least IMHO...

 

 

[Carlosan]

Activate beta tools in preferences to use the latest development in retopology

[AbnRanger]

4 hours ago, kenmo said:

I've been playing with InstantMeshes and for speed alone InstantMeshes blows the doors off of 3DC. It would be like a drag race between nitro burning GMC Supercharged 426 Hemi dragster and a tricycle.

Autopo has a long way to go to match InstantMeshes at least IMHO...

 

 

I didn't know we were drag-racing other Auto-Retopo tools outside 3D Coat. This tricycle seems fast enough for me, as long as I know how to use it properly and don't ask it to do more than it is designed to. If you give 3D Coat fairly simple objects to Auto-Retopologize, it does a bang up job, and does it in less than 30-60 sec. I could care less if another app does it in 15sec, if it doesn't match the result.

Sounds like Modeling in 3D Coat is not your cup of tea, and that's OK. Poly-modeling in other apps is just another approach or set of tools. Maybe using Blender + the HardOps addon is more to your liking?

[kenmo]

Blender is like Truespace or ZBrush to me. Terrible ui... :-)

3DCoat is very easy to get used to. The reason I purchased 3DCoat ver 3 way back... was for UV mapping and texture painting. UV Mapper Classic was cool but 3DC is too easy for UV mapping.  I love 3DC. Silo3D also has a retopology tools but no auto retopo.

For sculpting I still prefer Sculptris over 3DC for organics...BUT I'm more into vehicles.  So I really want to give 3DC a good try. I have diecast collection of  70+ 1/18 & 1/24 scale autos, hot rods, antique cars, muscle cars, etc... Love cars!!!! So, it's not the modeling in 3DC I dislike, it's the fact you must retopologize your model. As I said in another thread, Groboto3D converts to mesh in seconds.

Sorry any model I tried to convert in 3DC took a couple of hours. InstantMesh converted the same model in seconds. So it's not like comparing 15 seconds to 30 seconds. It's like comparing 3 hours to 45 seconds. Same with a trial of Zbrush I downloaded last month.

Don't believe me and I'll gladly upload my base mesh in the other thread for you to try an autopo on...

So I don't dislike modeling in 3DC. I like it very much. It's the retopo task which I really dislike. I realize this is not unique to 3DC and is common to sculpting apps.

 

Edited December 13, 2016 by kenmo

[AbnRanger]

Like I said before, 2 things you DO NOT want to try to use Auto=Retopo on.

1) Thin, single-sided meshes...like wings, automobile panels, or clothes.

2) Complex shapes

You can see at the end of the 4.7 video, that 3D Coat can handle quite a bit, but there is a certain threshold of complexity, that you don't want to cross. It takes some experience working with it to know where that threshold is. If your Auto=Retopo took more than a minute or so, then you took pushed it beyond that point. I'm thinking it had to be a thin, single sided mesh.

Again, if you keep your model separated into logical components and Auto-Retopo each of them separately, then it can work wonders. But if you try and merge a bunch of different shapes together and hand it off to Auto-Retopo, you will not like the result.

[AbnRanger]

 

 

[kenmo]

12 hours ago, AbnRanger said:

Like I said before, 2 things you DO NOT want to try to use Auto=Retopo on.

1) Thin, single-sided meshes...like wings, automobile panels, or clothes.

2) Complex shapes

You can see at the end of the 4.7 video, that 3D Coat can handle quite a bit, but there is a certain threshold of complexity, that you don't want to cross. It takes some experience working with it to know where that threshold is. If your Auto=Retopo took more than a minute or so, then you took pushed it beyond that point. I'm thinking it had to be a thin, single sided mesh.

Again, if you keep your model separated into logical components and Auto-Retopo each of them separately, then it can work wonders. But if you try and merge a bunch of different shapes together and hand it off to Auto-Retopo, you will not like the result.

As L'Ancien Regime posted above, Instant Mesh does not have the problems with complicated models that 3DC does. There is no need to dumb down your model when you use Instant Mesh.

And Instant Mesh was able to process a complicated model in the same time as 3DC can pocess a simple model.

If I use Instant Mesh there is no need for me to change my workflow to accommodate the 3DC's autopo. And I do model in a modular way.

If I am going to model simple models, then I'll use Hexagon3D, Silo3D or Wings3D.

I do not mean to argue with you AbnRanger as you are a wealth of knowledge however I feel you are completely missing my pooint...

Cheers & take care

 

[AbnRanger]

3 hours ago, kenmo said:

As L'Ancien Regime posted above, Instant Mesh does not have the problems with complicated models that 3DC does. There is no need to dumb down your model when you use Instant Mesh.

And Instant Mesh was able to process a complicated model in the same time as 3DC can pocess a simple model.

If I use Instant Mesh there is no need for me to change my workflow to accommodate the 3DC's autopo. And I do model in a modular way.

If I am going to model simple models, then I'll use Hexagon3D, Silo3D or Wings3D.

I do not mean to argue with you AbnRanger as you are a wealth of knowledge however I feel you are completely missing my pooint...

Cheers & take care

 

Why do you need Andrew to stop working on other needed parts of 3D Coat, to overhaul Auto-Retopo, when it already works pretty well and you have found a free remeshing tool that does an even better job...in your estimation? Sounds like you got what you want and there is no need to further debate this issue. It's not that 3D Coat cannot generate a nice quad mesh on complex objects. It's whether or not the topology is suitable for animation or further modeling. The videos I saw of Instant Mesh didn't seem to show a perfect Auto-Retopo solution. It has it's own limitations, and requires a good bit of clean up, afterwards. But if you just want a decent, non-animated quad mesh, 3D Coat can usually do that just fine.

 

I've Auto-Retopologized plenty of models in 3D Coat, complex and simple and it works pretty well for my purposes. It's not always perfect and hopefully Andrew can get around to improving it some more in the near future, but if I can't get the result I need in 3D Coat, it's good to know there are other solutions that might.

[kenmo]

I have no designs on being a pro 3D modeler, I don't do animation. I'm simply a hobbyist and old. I want to build props for my Vue renders. I'm 62 and would prefer to work more on the art side then on the technical side. And I am a technical person. I've been doing computer and network support (Novell Netware, SUSE Linux and now MS Windows server) for over 30 years. I build (or more accurately assemble) my home computers.

On my down time I wish to sped as little time as possible doing technical stuff. I find 3D modeling very creative and relaxing as I do  photography. I find retopologyzing more on the technical side and very frustrating. If I really want to do something hands on technical I'd rather be turning a wrench of my 40+ year old Chevy Corvette.

Thanks for the reply, I really do appreciate your thoughts and input.