Automate the UV Mapping Process

Finnn

Hello Polycount,

I noticed the amount of time I put into UV Mapping my 3D models (Max). I am really bothered with such an manual activity in an automated digital world that we (should) live in.
So me and a friend were wondering if we could make this world a better place and work on a automation tool for UV Mapping.
While researching I found some tools dedicated to unwrapping 3D models, but I havent tried any.

First of all, I wanted to get some opinion from you on these questions:

To what extent would you describe the UV Mapping process as automated?
Is the majority of artists using the native unwrapping tools, plugins or software dedicated only to this task?
How necessary will UV Unwrapping be in the future or will it be automated completely?
Is "Automated Unwrapping" a USP that could make this idea a vital concept for a standalone product?

ilari

To what extent would you describe the UV Mapping process as automated?

Not very, only packing algorithms have become really good. With simple meshes it's possible to get away with hard edge/curvature/cluster based unfolding but problems that sound trivial, like automatically differentiating a tube from a planar patch with a hole and marking a clean seam down it's length can actually be headscratchers. Deciding how islands should be cut up, which islands to rectify to the grid, figuring out a robust rectification algorithm, how to even begin handling special cases like trim UVs etc. are all tricky problems. 

Is the majority of artists using the native unwrapping tools, plugins or software dedicated only to this task?

I don't think I've ever actually seen a copy of RizomUV, Roadkill or such in the wild.

How necessary will UV Unwrapping be in the future or will it be automated completely?

Hopefully automated to a large degree.

Is "Automated Unwrapping" a USP that could make this idea a vital concept for a standalone product?

Definitely, especially if a one click seamless round trip between the UV-app and most common DCC software is available.

Eric Chadwick

You could start with seam generation as an initial task. That's one of the worst time-suck offenders that should be automated, imho.

The current crop of auto-unwrappers all tend to create lots of tiny islands, which pack poorly, and LOD horribly.

It seems to me this is an area of research that would be ripe for machine learning or artificial intelligence. 

wenglish

I've found my unwraps go a lot faster if I plan it while modeling. For instance, if there is a hard panel cut or sharp surface change, rather than just having 2 connected edges you can disconnect them and phase the edges through each other the UV "seam" is an inherent part of the mesh itself. 

Then I can just drop each polygon island in Headus UV and run "flatten" on each piece. So maybe in certain cases automation could be possible. You could at least automate the part where you hit "F" on everything 

RacePeaceDay

There are scripts (or feature included by default in some softare), make different UV islands from hard edge or smoothing groups with 1 click. Like "flatten by smoothing groups" in 3ds max. It's faster than manually splitting edges (only on cylinders need to split, because they're 1 smoothing group)

musashidan

Unwrapping today has come a looong way since I started doung it. So much so that it might as well be automated as it's so quick to do with the tools we have now.

Back then Max didn't even have an unwrap modifier and you had to make a face selection and stack shit loads of uvw mapping modifiers(one for every island) and mesh select mods using planar or cylindrical or whatever. Was a total nightmare in retrospect.

poopipe

The difficult bit is planning what you're going to do, not doing it.
The major packages all have decent tools now so with experience, the process shouldn't be a chore. 

Auto unwrappers are fairly mature now and will produce reasonable results very quickly but all fundamentally fail at perceptual interpretation of the mesh.
It'll get solved, it just hasn't been. 

With current tech... 
If you are short on time and have resources to spare then auto unwrappers can produce good results

If you value efficiency and quality then you need to take the time to plan and create uvs yourself.


Most game production will require that you manually uv most assets. 

Finnn

Thanks alot everybody for your input. Me and my friend discussed all your posts have brainstormed some more.  We also discussed the benefits machine learning could have on a project like that and had some ideas for use cases.  We also came up with an Plugin idea that shows the current optimization of the mesh for future uv mapping based on cleanness of the mesh, steepness of the edges etc

@Eric Chadwick - Seam generation would be a great start indeed. On hard edges the algorithm could easily detect good or necessary seams. But where seams are okay to be visible and where they shouldnt be is something alot more difficult to calculate...

Next semester we will have a course including 3D meshes (mathhh), procedural mesh generation and some more interesting stuff in OpenGL. I am very excited for that course!

RyanB

You could use machine learning for this.  Just need 10,000+ models as a data source.  A company could do this online.  You send them your model, they send you back an unwrapped model.  A big company like EA or Ubisoft has more than enough data to do this in-house.

poopipe

Establishing the metrics for what makes a good layout is the hard bit. You can't base it on coverage or texel density, you have to consider lodding and there's a whole load of stuff relating to shapes and good ways to unwrap them. 
It's a really tricky subject..