Blendshape Performance Question

[maulik13]

I am thinking to do something like this: create a duplicate of the face and rig that duplicate face with bones for squash and stretch and use this geometry as a blendshape target for the original geometry. I will always keep blendshape value 1 and then move the bones in the target shape that will reflect on to the original geometry.

However thinking of how blendshape would work I am wondering how this setup will affect on the blendshape defomer performance?

Is there a deformer which will directly copy changes from target mesh to original mesh whenever the target mesh changes? (eliminating any interpolation that blendshape deformer does)

Thanks.

[joojaa]

QUOTE(maulik13 @ 04/22/08, 01:32 PM) [snapback]284528[/snapback]
(eliminating any interpolation that blendshape deformer does)

Blendshape does not necceserily need to interpolate (wether or not any interpolation is calculated is counted in the overead of having a deformer). You can also just merge the polygons.

Or you can trasfer the rig in entirety so eliminate the blend shapes need altogether. Basically because you do not gain anything in particular with having the head separate except in modeling phase so you can always pre bake the step into the rig for animation. Now even if you need a proxy youve allready got one they dont actually need to be the same pipe as long as they are the same.

Also please note that just after skincluster node blendshape is the second most optimized node in maya, in 2008 ist fully multi threaded which means you have uite littel hope in beating it in performance even with a simpler node.

[maulik13]

Thanks Jooja for the detailed explanation. I appreciate it.

To elaborate more on what I was thinking is this. I read jason osipa's concept of global rig and local rig and I love that idea of modularizing the rig. So eventually my final head geometry (considering I am only rigging the head) or the global rig does not have anything except one blendshape modifier as the rig input. For my rig I am dividing it into different modules (local rigs) say, one is for lip sync, one is for emotions and one is for squash and stretch. Since I am doing all of this separately I have a freedom of adding more later. e.g. I can add more lip sync shapes later on if needed or if I need more control over the lips I can create a cluster based rig and attach it to the global rig. I think this is cleaner, manageable approach to rigging faces.

So since I have a hierarchy of blend shapes my concern was on how blendshape would cope with this. Especially since some of the blendshape targets will update live when I move clusters. So far I havent seen any major performance issues with blendshapes in such live update events.

I would love to hear on what do you think of such approach?

And good to hear about optimization of blendshape node. Sounds blendshape is the way to go.

[joojaa]

Yes i allways do a global and local rig. But i dont convolute the global rig by linking in the local, i replicate the local rig in the global one this way when the global is not visible the perfromance is same as with that approach, but when the global is visble and local is not i save some cycles.

In general sense it does not really matter but im a bit opposed on vasting cycles just because it takes the computer 1 minute more to do the job for you for permanent 0.002 seconds speed increase (now thet might sound low but 24 fps is 0,04 seconds so that is 5% of a frames time)

Tough today im converting to different method where the secondary rig i jut reading in the disk cache of the previous. this eliminates the need to compute the other rigs at all.

[maulik13]

So what happens if you have joints for eyelid and jaw on the face and on top of that you want to add squash and stretch using joints? How would you approach this?

Your disk cache idea is probably too advanced for me right now, if you dont mind could you explain it in a little simpler terms with more elaboration?

I appreciate you taking time to reply my questions.

[joojaa]

QUOTE(maulik13 @ 04/24/08, 06:54 AM) [snapback]284623[/snapback]
So what happens if you have joints for eyelid and jaw on the face and on top of that you want to add squash and stretch using joints? How would you approach this?

The workflow is the same i jut eliminate redundant parts in the netwrok by not working then in together

CODE
node -> node -> shape \
                       blend -> next step -> shape
                shape /

is equal to:

node-> node ->shape and

node-> node ->next step ->shape

Now nothing says the inputs of node node cant come from same source or in case of cluster s that they aren't shared. Note i save at least 3 nodes computation time. While not a big thing here ist still time thet cane be saved without impacting ANYTHING in the process of things. (the data comes in from a character set anyways)

There are however places where the blend trick is imperative. because you really want to blend the data in. tough nothimg wrong with having reservation for it.

Its the sort of all needs to be in one networks thinking that blinds people to facts.