Question concerning SVN log messages

This forum is aimed at user contributions, in the form of assets, side projects, code patches and similar.

Moderator: joepal

Question concerning SVN log messages

Postby Solkar » Wed May 08, 2013 6:38 pm

To "suggest and contribute" it's good to clearly understand recent code changes, thus I start a thread for that purpose.

------------------------------------------------------------------------
r5084 | thomas.larssonXXXXXXXXXX | 2013-05-06 12:13:31 +0200 (Mon, 06 May 2013) | 3 lines
[...]On most clothes the body shapekeys are empty. This is not a bug in the program code, but because the shapekeys do not extend to the helper geometry. Since the rigs will be reconstructed for alpha 8[...]
(emphasis mine, email address masked)

Which rigs, pls?

------------------------------------------------------------------------
r5085 | jonas.hauquierXXXXXXXXXX | 2013-05-06 14:31:58 +0200 (Mon, 06 May 2013) | 3 lines
[...]Rotation angle compensation for target rigs still to do.
(email address masked)

Is that about issues like e.g. Leg deform?

Regards, S.
User avatar
Solkar
 
Posts: 63
Joined: Thu May 02, 2013 5:34 pm

Re: Question concerning SVN log messages

Postby ThomasL » Thu May 09, 2013 3:14 pm

The comment means the following. If you import a character with clothes (here the benchmark tights), and pose the arm and shoulder up, pure bone deformation gives the result in the lower picture. If the character was exported with Body shapes selected, there is a corrective shapekey that improves deformation. As we see in the upper picture, the shapekey does not extend to the clothing. The reason for this defect is not in the program code, but in the data file defining the shapekey. The shape is only defined for the body, and not for the helper geometry, hence the difference between the body and clothing deformation.

arm-up.png

Here is the driver for the shapekey in question.

arm-up-driver.png
ThomasL
 
Posts: 1139
Joined: Tue Sep 15, 2009 2:46 am

Re: Question concerning SVN log messages

Postby Solkar » Thu May 09, 2013 4:09 pm

Thx for your efforts, Thomas, but what you explained is the part of your message I had understood.
I wanted to know which rigs will to be "reconstructed".

Regards, S.
User avatar
Solkar
 
Posts: 63
Joined: Thu May 02, 2013 5:34 pm

Re: Question concerning SVN log messages

Postby ThomasL » Fri May 10, 2013 3:11 am

The new mesh will be in a relaxed pose with bent joints. This will affect all rigs to some extent, due to the need for non-trivial roll angles. Most joints have a preferred bend direction, and roll should be adjusted so this is either local x or z.

The MHX rig depends on the rest pose in several other ways as well, and I'm not sure exactly what changes need to be made. One can hope that it is possible to do most of the deformation with bones alone. E.g., the pose above could be improved with a deltoid bone stretching to the upper arm. In T-pose this leads to problems when the arm bends down, but if the arm is already down in rest pose a deltoid should work better.

I believe that corrective shapekeys will remain necessary for best results, but they will have to be remodelled from scratch. Therefore it does not seem worthwhile spending time on targets for the current mesh that will be replaced very soon.
ThomasL
 
Posts: 1139
Joined: Tue Sep 15, 2009 2:46 am

Re: Question concerning SVN log messages

Postby Solkar » Fri May 10, 2013 6:03 am

Thank you very much, Thomas,

that's what I needed to understand.

ThomasL wrote:One can hope that it is possible to do most of the deformation with bones alone. E.g., the pose above could be improved with a deltoid bone stretching to the upper arm. In T-pose this leads to problems when the arm bends down, but if the arm is already down in rest pose a deltoid should work better.


I think considerations like these are in the vicinity of the scalar rigging¹ issue I mentioned, and afai see any solution meant to maintain interoperability with industry standard animation systems has to deal with that; and of course esp those with the stricter constraint of COLLADA / X3D (hanim) "encodability".

The two usual ways to deal with that
- "synthetic" bones, like the "deltoid bone" you propose
- and "event" based morph correction, like e.g. made available by (esp. bone-)driving blender's shape-keys,
none of them being overly attractive.

The other way is what animorph did - using a proprietary pose system at the price of giving up interoperability; esp with IK-solutions.

---

Summing up - there is an issue.

===

But this
ThomasL wrote:This will affect all rigs to some extent, due to the need for non-trivial roll angles.
(emphasis mine)
to me appears a sort of brute force refactoring - I don't need a crystal ball to predict that this will trigger changes to the mesh itself, which may have side-effects which may trigger further changes and... - you get the idea.

I've followed the makehuman development sporadically for quite some time now and I sometimes had had some doubts if the project would get to a point which justifies reconsidering vertical ranges of production for professional usage.

That point imo appears now in reach - excellent job by you and your team, btw!

Thus I'd like to reiterate my idea about a sort-of "dual" solution - only minor fixes within current conceptual bandwidth not risking made milestones and conceptual innovations only by means of the "extra" paradigm I tried to sketch in the aforementioned thread.

---

The other thing I'd like to suggest is a sort of adaption of the concept of "effective theories" in science:
Regardless of how a certain final pose state will be achieved - there is e.g. sth like a (however weighted, however posed) lower leg, and that certain leg has a (however parttitioned) surface and a certain volume.
viewtopic.php?p=19599#p19599
Solkar wrote:I'm wondering if, because of the very simple form of the backside of the human leg from just underneath buttock to just above ankles, one could try surface fitting that with a low-order polynomial and work with a bit of diff'geo and measure to get a correct volume-preserving morph.


Those real constraints given, one can try and deduce whatever model-specific realization of those constraints.
That way one could get a collection of model-independent assets which would survive even the most crude refactoring needs, or am I mistaken?


Best regards, S.


¹I'n not "married" to that phrase, but I wanted to coin one.
User avatar
Solkar
 
Posts: 63
Joined: Thu May 02, 2013 5:34 pm

Re: Question concerning SVN log messages

Postby duststorm » Fri May 10, 2013 10:15 am

r5085 | jonas | 2013-05-06 14:31:58 +0200 (Mon, 06 May 2013) | 3 lines
[...]Rotation angle compensation for target rigs still to do.


That commit has nothing to do with the blender export. It's for loading BVH data from within MH.

Maybe for the new basemesh we can use a different approach for retargeting, without manual roll/rotation compensation, and only require a mapping.
For example this: http://infoscience.epfl.ch/record/12139 ... _EG_00.pdf
or http://www.arishapiro.com/mig12/mig12system.pdf
MakeHuman™ developer
User avatar
duststorm
 
Posts: 2566
Joined: Fri Jan 27, 2012 11:57 am
Location: Belgium

Re: Question concerning SVN log messages

Postby dmbasso » Fri May 10, 2013 10:29 am

Yesterday I've committed the addition of a MakeHuman retargeter to my CMU Mocap Library Browser addon for Blender. It is more of a small hack, but for those mocaps it works relatively well, although there is plenty of room for improvements. In the afternoon I'll post in the gallery a video of some motions applied to a MakeHuman rig.
dmbasso
 
Posts: 21
Joined: Sat Apr 30, 2011 3:06 pm

Re: Question concerning SVN log messages

Postby ThomasL » Sat May 11, 2013 6:55 am

Solkar,

My own main interest is to export a rigged mesh to Blender, and hence I am limited to the tools available there. I'm not sure what the animorph stuff you talking about is - that apparently happened before I joined MH - but in general I doubt that many people would use it. MH's strength is as a content creation tool, not as another posing tool.

I think I know the solution to the roll angles now. Since the joints are slightly bent in the rest pose, a bone and its parent define the bending plane, and you can figure out the roll angle from that.
ThomasL
 
Posts: 1139
Joined: Tue Sep 15, 2009 2:46 am

Re: Question concerning SVN log messages

Postby Solkar » Sat May 11, 2013 11:58 am

ThomasL wrote:I'm not sure what the animorph stuff you talking about is

The (C++) morph- and pose-backend of mh 0.9.1.

ThomasL wrote:My own main interest is to export a rigged mesh to Blender, and hence I am limited to the tools available there.

Mine as well, and if it comes to blender, the combi shape keys + driver is a totally sound concept, so there would be no need at all to replace that with unreal bones like the "deltoid bone" you propose.

What I'm talking about is backing the creation of whatever respective concept-related functors by as-widely-as-can model-independent means, for detail pls. see the aforementioned "Leg deform" thread.

---

To make it a little less abstract - yesterday I've learned that indeed a base-mesh re-factoring is going on.
That currently brings it with it, that all work spent on the development of the current shape keys for shoulder correctives will be made obsolete at a single strike.

Although there is a modern programming paradigm to re-factor generously, this is not imperative for every situation.

The current methodical problem is that you re-factor at two fundamental components of the system
- the base-mesh itself
- and the pose and morph concepts.

That takes you widely out of well-tested grounds.

Regards, S.
User avatar
Solkar
 
Posts: 63
Joined: Thu May 02, 2013 5:34 pm

Re: Question concerning SVN log messages

Postby Solkar » Sat May 11, 2013 1:42 pm

@Moderators:
My initial reason to start this thread was to gather explanations about recent SVN log messages, the respctive questions have been kindly answered in-depth.
Because the current discussion of Thomas' and mine carries us far beyond that SVN-log-related purpose, I propose splitting this off into a new thread, if that's o.k. for Thomas.
User avatar
Solkar
 
Posts: 63
Joined: Thu May 02, 2013 5:34 pm

Next

Return to User contributions

Who is online

Users browsing this forum: No registered users and 1 guest