V4.1 BETA (experimental 4.1.17D)

[AbnRanger]

I can say at this point, that the improvements to Auto-Retopo are both a Marvel and a Misery. At times it can do such a good job, that you're blown away. And other times, it just bogs down over and over....even after multiple attempts and tips are followed. The whole point behind Auto-Retopology is to make the job easier and faster....not such a hit and miss prospect that you could easily end up wasting far more time than you ever bargained for. I've blown enough time trying different settings (3 full days) that I could have manually retopo'ed a handful of characters.

 

Can't recommend using it at all, with this in mind. Something has to be done about the sheer volatility and wasted time, or I'd say just scrap it and go back to the old engine. Better to just spend the time improving the manual retopology tools and moving on from Autopo, the way Andrew has the Tweak Room.

 

There are still some feature requests that have been long requested and sorely needed...such as the need for a shell modifier...which would be nothing more than having an option with NORMAL EXTRUDE....to simply extrude in both/all directions.

 

http://www.youtube.com/watch?v=lCdrU_vsLA0&feature=youtu.be

 

Auto-Retopo did a good job on the body of this Banshee,  but it was/is HELL trying to get it to work on the rest.

http://www.youtube.com/watch?v=lCdrU_vsLA0&feature=youtu.be

 

http://www.youtube.com/watch?v=H3jwEx2sxWI

 

EDIT: I went back into this forum to find something Andrew stated about voxelizing (a copy of) the model first > resampling to 300-400k > SMOOTH all > Decimate in the Dialog. So, it's better, but having to reduce the model so much beforehand, destroys areas like eyelids and such....not good. It needs to be able to work with higher poly counts and not bog down.

 

 

Edited October 3, 2013 by AbnRanger

[BeatKitano]

You're just thinking this is a solution for mesh output, it's not, won't. Zbrush didn't invest heavily in this tech with verold to create perfect output meshes. They went the intermediary stage: offer a quick "cleanup" process to help sanitize the freeform mesh. That's it. 

I could go "I told you so" a few pages back, but that would just be arrogant, and to be honest the tech is not bad it's just not used in a smart way, without a few addition (reproject) this tech offers little advantages to 3dcoat as the mesh is not animation friendly AND can't really be used in the production process as 3dcoat is not really relying on quad based meshes.

What started as a fancy marketing selling point and somewhat quickening process in the first iteration is now an obsolete function serving little purpose to 3dcoat freeform paradigm, it's here Andrew has to play his cards to keep the upper hand, otherwise it's just a "trying to catchup" strategy and that doesn't do any good if the result is not better than the competition.

 

Now I'll just shut up since this kind of discussions aren't allowed here (while they may be the most insightful one for Andrew from artist's pov). 

[AbnRanger]

You're just thinking this is a solution for mesh output, it's not, won't. Zbrush didn't invest heavily in this tech with verold to create perfect output meshes. They went the intermediary stage: offer a quick "cleanup" process to help sanitize the freeform mesh. That's it. 

I could go "I told you so" a few pages back, but that would just be arrogant, and to be honest the tech is not bad it's just not used in a smart way, without a few addition (reproject) this tech offers little advantages to 3dcoat as the mesh is not animation friendly AND can't really be used in the production process as 3dcoat is not really relying on quad based meshes.

What started as a fancy marketing selling point and somewhat quickening process in the first iteration is now an obsolete function serving little purpose to 3dcoat freeform paradigm, it's here Andrew has to play his cards to keep the upper hand, otherwise it's just a "trying to catchup" strategy and that doesn't do any good if the result is not better than the competition.

 

Now I'll just shut up since this kind of discussions aren't allowed here (while they may be the most insightful one for Andrew from artist's pov). 

I just think the two are coming at it in separate ways and I don't see a single thing wrong with Andrew's approach. 3D Coat does not NEED an intermediate, cleanup reprojection tool. Perhaps that is too engrained in your membrane (after using ZBrush for so long?), that you just can't see any other way than what Pixologic dictates. I can see why it would be important for Zbrush's quad-based approach, but trying to make Autopo Retopo fit a mid-stream sculpting workflow in 3D Coat isn't getting it.

 

With voxels, it remeshes constantly. Auto Retopo was always an OUTPUT mechanism, and it works well in that capacity...for the most part.

 

http://www.youtube.com/watch?v=_aMkWrnl4ds&feature=youtu.be

Edited October 3, 2013 by AbnRanger

[BeatKitano]

Voxels are slow, they are uniform, you need enormous quantity of ram to get a decent res for crisp details.

Don't give the voxel argument it's kind of silly (unless Andrew take the road all the way and go with SVO).

 

If you're cool getting stuck with a destructive one way step down process (decimate/reduce) to go back in your detailing process with liveclay, fine ! I don't.

 

And it's not about habit it's about practical use in a production environment. 

Photoshop gave layers so that you can make quick changes without doing everything all over again. 3dcoat has no layer, reprojection would be a way to make your changes without doing everything all over again.

Andrew gave autoretopo another use case, but he doesn't go all the way (again ?).

[gbball]

Is Andrew done working on the autotopo routine?

 

I thought it was a work in progress still.  A lot of people here, including me said he should take the time to get it right, rather than going part way and then feeling pressure to work on something else.

 

I still think it's potentially a very valuable feature, that will allow 3D Coat to integrate better into a pipeline with other software.  However, (and this is the most critical part) We need support for quad based meshes in the sculpt room, this will make Autopo more worthwhile, because we can covert to quads and continue to sculpt...there are also cases where we bring in elements from another program to add to something we are sculpting.

 

If we are just creating characters in 3d coat, then it doesn't really matter, we can sculpt it however we want and then, retopo it manually or automatically.

 

However, if we model something in Maya for example and apply a level 2 subDivision surface to it, then bring it into 3D-coat to add it to our character sculpt (say a vest or hat), then we automatically lose all that work we did, because we have to either voxelize it or triangulate the mesh.  Why can't we have the option of bringing it into sculpt room as is?

 

The image below is a real example of what I experienced using 3D Coat in production.

 

In Blender, I modeled a few elements, a couple flowers, and a hexagonal honeycomb pattern to use as a curve profile.  I made everything with a low poly quad based mesh, then applied a few subdivision levels to smooth it out.  

I made the flowers merge objects so that I could easily position them using the onpen feature of the merge tool.  This helped me get them onto the landscape that I built in 3D Coat in the way that I wanted very quickly.  I also used the Curve tool to position the honeycomb pattern in a nice curve to frame the set.

 

This works great as a previs tool, since I'm not technically a modeler at work, I'm a designer.  It allows me to quickly make sure all the shots will work and stuff like that for the show.

 

However, upon getting the concept I created approved...I basically have to tell the modeling team to redo everything from scratch...since the work that I'm able to then export from 3D Coat is essentially unusable to them, aside from scale and positioning information.  I'm not going to retopologize everything and export it since that's not my job and it would take too much time.  So what I end up doing is giving them a decimated obj of the set I made along with any objects that I first modeled in Blender and then have them recreate the set, based on what I did.

 

How much of a time and money saver it would be if I could just reexport the work the same way I brought it in.

 

I understand the freesculpting philosophy of 3D Coat, but at some point, if you want to integrate it into your pipeline, you'll want a more nondestructive workflow.  

Imagine having a library of mergeable meshes that are all in their orginal quad based mesh...I know it's possible for meshes to be saved as quads in the merge library, but why no support for them in the sculpt room?

 

Edited October 3, 2013 by gbball

[manticor]

Im hoping he is still working on it .He's past the point of no return so to speak ,we all can see that it can produce really really good results ,it's just the consistency ,speed and stability need to be improved .

[AbnRanger]

I don't feel nearly as ill about it today as I did yesterday. It's just that it seems way too finicky and requires too many workarounds to just work. If that can be resolved, it will be truly impressive.

[Carlosan]

Andrew and Raul are merging the Raul's branch code to the Andrew's trunk code.

 

This process take time to test.

[manticor]

I don't feel nearly as ill about it today as I did yesterday. It's just that it seems way too finicky and requires too many workarounds to just work. If that can be resolved, it will be truly impressive.

To be totally honest after using zremesher for while now the same can be said for that too,while it does an impressive job overall it does take good while to get something that you want if you are making game res models .its still a test , undo ,add curves ,test ,undo paint colours ,test kind of workflow and although it is fast at generating ,you end up sitting there for ages till you've got something perfect.

So now my workflow is to go with 25 - 50 % extra polys and optimise down ....which is a problem with zremesher as it generates spirals almost every time .

Andrews algorithm seems pretty good at getting a clean mesh with a good flow which you can take out edge loops easily.the only thing that seems to let it down other than the points that have been mentioned is it hugging crevices and lateral twists of loop flow on areas like fingers .

What I think would be nice would be a dot loop range select option like what abn suggested and a kind of shrink wrap magnet that pulls points towards creviced areas ....kind of like baking a curvature map in xnormal and using this as a weight map to drag points towards the creviced areas. So you could use the autopo to go higher than you need to get good detail capture , quickly hand optimise pullin g loops out and then tighten up the final mesh.

[mercy]

Andrew and Raul are merging the Raul's branch code to the Andrew's trunk code.

 

This process take time to test.

What? Raul is working with Andrew actively now? In realtime, like in an office or through skype? WOW!

[Carlosan]

What? Raul is working with Andrew actively now? In realtime, like in an office or through skype? WOW!

 

http://3d-coat.com/forum/index.php?showtopic=7912&p=100393

 

Hi everyone!!!!!!

 

 

yes, I'm back again, not permanently but definitely I'm happy to be again here!!!

lots of exiting stuffs ahead!

[Andrew Shpagin]

Seems I achieved stage when I satisfied with autoretopo result. I will do beta update very soon, possibly with Raul's changes (but not sure).

As always, at the very final stage I got idea how to do all in very other way and much better But will not go that way right now. Possibly I will ask Raul or someone else to implement new idea.

I will get back to general bugfixing asap. My roadmap is to get clan very clean build, call it 4.1 and then move further with requests that are well known.

 

Thanks for patience and support in this not easy time!

[philnolan3d]

One thing that's been bothering me on my current project is that my model is all very thin parts and I'm having a lot of trouble working on one side without it messing up the other side. For example with smoothing or using CleanClay on just one side. it'd be nice to see Ignore Back Faces with brush tools.  The Lute tool seems to work just fine on one side without affecting the other.

[artman]

 

As always, at the very final stage I got idea how to do all in very other way and much better

hahaha.. Maybe if you didnt went trough that whole process you wouldn't have found new idea.

[Carlosan]

Greats news ! Ty !

[AbnRanger]

One thing that's been bothering me on my current project is that my model is all very thin parts and I'm having a lot of trouble working on one side without it messing up the other side. For example with smoothing or using CleanClay on just one side. it'd be nice to see Ignore Back Faces with brush tools.  The Lute tool seems to work just fine on one side without affecting the other.

There is an IGNORE BACKFACES option...in the E-Panel (lower left corner)

[philnolan3d]

There is an IGNORE BACKFACES option...in the E-Panel (lower left corner)

 

That only seems to work with the "lasso" types of selection. I knew about it and have it checked.

[AbnRanger]

That only seems to work with the "lasso" types of selection. I knew about it and have it checked.

The only time I've had issues with something being affected on the opposite side is in the Retopo Room, when using the Strokes tool or working on a thin, double-sided object.

[PolyHertz]

I will get back to general bugfixing asap. My roadmap is to get clan very clean build, call it 4.1 and then move further with requests that are well known.

 

Looking forward to it! Requested features more so, but there are a few bugs that would like to see squashed as well, such as:

 

http://3d-coat.com/mantis/view.php?id=1132

http://3d-coat.com/mantis/view.php?id=980

 

There's also a bug where the paint layer opacity slider floods the undo stack while being dragged. I thought it had been reported a while ago but not running across it atm.

[geo_n]

Testing the new autoretopo.

How to avoid 3dc getting stuck during the process? It gets stuck at 88% consistently.

[TimmyZDesign]

Testing the new autoretopo.

How to avoid 3dc getting stuck during the process? It gets stuck at 88% consistently.

I don't think this current available version (as of right now) is the version that Andrew is "satisfied" with. Wait a little bit longer and he will upload the new and improved one soon. (see post #2803 above)

[geo_n]

Ok thanks. I'm liking the results so far when it does work. The edgeloops and polycount density is so much improved than previous autoretopo.

[Andrew Shpagin]

Updated to 4.0.08 (Win)

- changes mostly related to quadrangulation, now it should be more stable and predictable.

[Carlosan]

ty

 

testing !

[L'Ancien Regime]

Downloading...

[artman]

Im getting same result as older build,only calculation takes longer time...sorry.

Horn is still not captured at all....Im wondering if you even  tested with this file I sent you.

(I did not use guide as it is the reason for the test, I should not need to use any for such simple object)

 

I still dont think all this work is time lost since it helped you discover new method.

So I think I'll wait until you implement new idea someday and stick to manual retopo for the time being.

[artman]

here is result with guides (very few non contradicting guides (3-4))

 

I get 2 disconnected objects and holes...

[Andrew Shpagin]

Im getting same result as older build,only calculation takes longer time...sorry.

Horn is still not captured at all....Im wondering if you even  tested with this file I sent you.

(I did not use guide as it is the reason for the test, I should not need to use any for such simple object)

 

I still dont think all this work is time lost since it helped you discover new method.

So I think I'll wait until you implement new idea someday and stick to manual retopo for the time being.

autopo horn.jpg

Excuse, seems I missed file and can't find it now. Please send again.

Have you tried without guides? And try bigger polycount.

[artman]

first image is without guide (as I wrote)

 

 

File can still be found on this page ,its very small 6mb file made for testing purposes.

http://3d-coat.com/forum/index.php?showtopic=10395&page=139

 

Im not gonna try above 2000 polys as it is the very purpose of this test.

2000 polys should be enough

(zremesher get it nice one shot...I dont want to annoy you with zb comparison but its the only reference I've got to compare )

At 2000 polys I do not expect to get arrows captured  but I expect the horn to be captured,and no holes at least.

If I need 5000 polys and more to get this simple small object a decent base, I just would use  manual retopology instead.

 

Edit:Im using a 1000 polys witch produce 2000tris.

 if I use 2000polys(which gives result of approx 5000 tris)  result is better but 5000 tris is too much.

[Andrew Shpagin]

first image is without guide (as I wrote)

 

 

File can still be found on this page ,its very small 6mb file made for testing purposes.

http://3d-coat.com/forum/index.php?showtopic=10395&page=139

 

Im not gonna try above 2000 polys as it is the very purpose of this test.

2000 polys should be enough

(zremesher get it nice one shot...I dont want to annoy you with zb comparison but its the only reference I've got to compare )

At 2000 polys I do not expect to get arrows captured  but I expect the horn to be captured,and no holes at least.

If I need 5000 polys and more to get this simple small object a decent base, I just would use  manual retopology instead.

 

Edit:Im using a 1000 polys witch produce 2000tris.

 if I use 2000polys(which gives result of approx 5000 tris)  result is better but 5000 tris is too much.

I have not found file there.