Thanks Thornae. Looks very interesting. As a coincidence, Neablo just sent me that same link.
Argh, I have to go to a pc running windows. I can’t see any of the movies.
Edit: Awesome! I wish I had animation files to test my own model.
I managed to get to the movies:
http://staff.ci.qut.edu.au/~barkerc/Final%20PAN%20website/movies/
http://staff.ci.qut.edu.au/~barkerc/Final%20PAN%20website/movs/
Oh no, you understood me wrong, toloban!
This was no coincidence I sent you the link because I worried you wouldn’t see it here.
The credits all go to thornae! A very cool link, indeed
Edit: Awesome! I wish I had animation files to test my own model.
What do you mean?
Motion capture.
I just thought of something and figured this is the best place to post it. Is there a way to have a shape key activated by a combination of roll angles? I mean, a shape key is only activated with a combination of say x, y axis instead of just x-axis. If so I would imagine proper deformation of the shoulder would be much easier.
Another explanation:
shape key is driven only when two roll angles are activated. So for rotations of x=0 the key is not activated. If y=o the key is not activated. If x=30degrees and y=30 the key is = to 1:
x ------y ------key strength
0 ------0 -----------0
0 -----30 -----------0
30 -----0 -----------0
30 ----30 -----------1
15 ----15 ----------.5
From Python, a driver can use any combination of values, sure. But I try to imagine what shape key I would want to make that would work when a bone is rotated on two axes with equal value. How could that be easier than binding four shapes each to a simple position: extend forwards, backwards, up and down? All in between is blending those shapes.
By the way, I think roll angle refers only to the Y axis.
Hi Toloban,first all,good work,really interesting.
For proper shoulder deformation the thing to do in MHO isn’t creating a set of shape for every x,y,z rotation but it’s to rig the shoulder in a way that we can measure the global orientation of a pose,like measuring the dot product between 2 bone that point one in the rest pose of the joint and the other in the deltas we are interested in(the shape at a particular joint space location).In this way we can effectively know when using a that shape without having many problems.The roll value should be keeped separated.
This approach is the same made in the ILM hulk film.What I find really problematic is modelling the pose with Blender crazy space,it’s really disturbing and difficoult(some time ago I was trying to do a pose space deformation scripts but I don’t have time,the idea was to copy the mesh in the pose we want to fix,reverse engieenering the rotation and after reapplyng the delta as a shape key)
As i understand you mean to create shapes for certain key poses (“deltas we are interested in”) not related to what we know as XYZ. I only wonder if blending those keys with each other wouldn’t in the end become too difficult to make it worthwile.
Maybe more than the 6 basic XYZ shapes (X+,X-,Z+,Z-,Y+,Y-) I am using now would be necessary. At least for this model those basic shapes have worked unexpectedly well, with only one pose that would need some tweaking, but being an unusual and extreme pose, I can leave it as it is.
That script would really be most useful! Is it really too difficult? Can any of the exporters bake the mesh to its deformed state to understand the “forward” process? I don’t complain much. I have got used to crazy space, my only regret is that working like that is too time consuming.
That is not likely to change soon, as Ton Rosendaal explained here:
http://projects.blender.org/tracker/?func=detail&atid=264&aid=5078&group_id=9
HI Toloban,more than 6 basic shape are needed.Think,for every joint(for now the shoulder)a unit sphere centered at the joint pivot.This become our joint space.Every possible orientation of the shoulder is on this unit sphere(but the roll need special attention).
If we organize in a matrix these positions we can find the right shape(simply tracking some vector.The big problem is finding a good rig solution which made things not cumbersome).
About the crazy space,I think ton missed the point.It’s not important doing in realtime as now,it’s only important that one can find the difference between a deformed mesh which need fixing and a mesh deformed and fixed(and this can take some time(seconds),it’s depend on the implementation).We must work in world space,and after having found the deltas we must put them in object space(as a ordinary shape).The reverse enginerieng of the bone deformation in this way can be done(I have some code which is able to find how much every vertex is trasformed,what I don’t have done is the reapplication of the delta to the object,but this is an old project of mine,never completed )
About the difficoult in blending shape,for this we’ll need a new type of morphing,not additive,and probably it’s not a good thing doing it in python.
Renderdemon, do you mean it should be possible to edit a deformed mesh in normal 3D viewport space, with all the regular tools available (proportional editing, dragging constrained to an axis, widgets) if the actual location calculation is done afterwards? Very often the “small steps” Ton suggests are not possible. With very bent bones it’s madness.
Too bad I can’t really help to find a way to apply the reversed deformation, as an alternative to a crazy space fix.
About creating keys for poses that would be triggered by the way a vector points, you could use pydrivers. They can use any vector or matrix you want.
But since by using standard weight painting you still have to deal the crazy space problem, here is a crazy idea: why not use modeled poses as shapes driven by an armature which has no weight influences on the mesh? In the end it would feel like you are using a normal armature, you could even add IK solvers and all that stuff.
(Even though this sounds interesting I don’t think I am going to dump my current six shapes approach, distortions, pinching and artifacts are quite elliminated after all the necessary corrections)
Yes Toloban,it should be possible editing the mesh with all the tools we usually use in modelling(now probably even with the sculpting tool!)
The reason that works is because I don’t try to do the inverse trasformation(which can’t exist)but I simply construct,for every vertex(after is trasformed in world space),a local frame that store the deltas(multipling with the inverse matrix of this matrix should give the right values).
From a mathematical point of view,I have the problem solved(probably there are some minor issue that can happen but the method should work,at least for my knowledge),but it’s an old project I have started on may/june and my attention has gone or other stuff recently(normal maps,sss,sculpting,ecc.).The major problem which made me stop was how to do all the stuff,because for every joint we need to save where(in joint space)the deformation occur,we have to copy the mesh,modify the mesh and after calculate the deltas,that finally get saved in the shape key.I’m not really a coder and creating a good script that does all this stuff is a lot of work,I should had to study how events work,or make different scripts for the various task involved,but I’m lazy.
Btw,you are doing a really good work,the best I have seen here in the blender community
Bye
Well, your animations are so great. :yes: They’re very lively and helpful in the real life and in medicine.
Thank you for letting me know, please feel free to email me if problems occur in the future.
[url=http://simulationcredit1.com]simulation credit[/url
