Idea for a tool : tweak weights

[rray]

hi,
i was thinking it would be interesting to have a move point tool that instead of moving the point by modifying the base mesh, does so by influencing its envelope weights. This is supposed to be an alternative to weight painting, but you would not have to choose which weights get modified. the tool would pick an optimal combination of weights so the target point can be reached. The interaction would be identical to the move point tool.

This would of course not work at all on the model in its base pose and also often the targeted point is just not reachable by modifying weights . then it could just try to reach a close position. additional bones would have to be added by the user then.

... just putting it out there maybe this exists already, or ideas what math tricks/algorithms could apply here.
wondering what is out there already concerning this
-rr

[Sil3]

Looks like a promising tool , specially because I personally have a hard time using the Smooth Weights in Softimage, having the need to first Select all the point I don't want smoothed, Lock their weights and Smooth, then Unlocking everything again, very tedious for me.

On a side note but pertinent to Envelope Weighting, I come across this free tool for Maya and I also would love to have this on Softimage


ngSkinTools



Is this even possible to code for Softimage?

[rray]

Do you mean selection masking like it exists in vertex color painting? Yep that would be cool. Overall I think the current Softimage weighting toolset is really good, but the brushing feedback could use an overhaul.

Maybe to simplify this imaginary project, instead of a tool, one could do some secondary shape modeling, and call a function "approximate shape with evelope weights".

Still a big task maths wise, are there any algorithms which optimize a set of values to reach a certain target value? neural nets?

[OzAdi]

Looks like a promising tool , specially because I personally have a hard time using the Smooth Weights in Softimage, having the need to first Select all the point I don't want smoothed, Lock their weights and Smooth, then Unlocking everything again, very tedious for me.

Am I missing something?
why would you select the points you dont want to smooth, lock them, then smooth? wouldnt it be easier to just select the points you do want to smooth, and hit the smooth button?

[Sil3]

Yes that what I thought at first... but it didn't worked, Softimage would also start scrambling the already painted weights on other areas by redistributing the smooth influences... this was in 2013, I haven't used Softimage since last year, im now in Maya land, fortunately only as an animator or else I would do crazy

[OzAdi]

I think this is because normalization of weights is on, I am not in front of soft at the moment:)

[Mathaeus]

hi,
i was thinking it would be interesting to have a move point tool that instead of moving the point by modifying the base mesh, does so by influencing its envelope weights. This is supposed to be an alternative to weight painting, but you would not have to choose which weights get modified. the tool would pick an optimal combination of weights so the target point can be reached. The interaction would be identical to the move point tool.

This would of course not work at all on the model in its base pose and also often the targeted point is just not reachable by modifying weights . then it could just try to reach a close position. additional bones would have to be added by the user then.

... just putting it out there maybe this exists already, or ideas what math tricks/algorithms could apply here.
wondering what is out there already concerning this
-rr

Yeah, something like this, I think it definitively wants something else, instead of linear blend. As linear blend simply can't describe the "outer" side ( elbow, knee, so on), outside of possible blend area. Maybe some negative blend. don't know...

BTW there is a number of "advanced" systems for Max for a long time, like very old Physique enveloping operator, providing spline based envelope, bulges.. but nothing as you described - as far as I know. Generally I think this Physique thing was considered as overcomplicated for average user.

Personally, had a bit different idea, to create completely procedural set of deformers, all together acting as an unique "envelope". Instead of bunch of bones for roll division, just simply twisting, which can't loose volume. Instead of intermediate bones on 2d joint, rotation around joint for outer area, shearing for inner area. Small changes are performed with kind of corrective shape keys in another, previous operator, in this scenario. That's because I think, only a huge difference against traditional weighting, could motivate people to use the new system. Also, machine is able to do a very precise sorting at small distances, instead of manual weighting by humans. Otherwise, some 20 -30 % improvement, together with load of new, specific rules, I don't know....
But never had enough motivation to create complete system. Also in case of ICE, it somewhat tedious to do the caching system for static values. Anyway that's how last version of 'rotation for outer side, shearing for inner' looks like. It has a simple "collision avoidance" too, actual just simple clamp to plane.

[rray]

Interesting, there is a need for an industry standard for more advanced enveloping methods other than linear blend. Problem is that it can get incredibly complicated very fast so simplicity of use is easily destoyed.
Dual quaternion is a good first step and solves some problems (btw I think it's impressive but bad naming, it should be "radial blending", as opposed to "linear blending", which describes what it does)
The tool/plugin could support the DQ weight as well. It's just an extra parameter to consider.

[SpookyMunky]

Personally I'd be hard pushed to take one of these "advanced" automatic weight systems and then hand it on to an animator hehe. Until it gets to the point where the entire weighting is 100% spot on, the actual process of going through it and checking every point in my mind takes as long as it would starting from scratch ? I'd be more interested in a system that just mirrors weights properly as you work, remove the need for mirroring.. if we are daydreaming about new stuff....

Edit : in reality (again imho) one way you can speed up weighting is to create a simple purely volume character for any given rig, e.g. for gear create a fat(ish) character and weight it properly, for each new character you want to weight for that rig, load up fatty and use either its own rig / move the points around by hand with nice big linear falloffs until it matches the current character roughly, gator new mesh to that, clean it up........ not a very elogant solution but could work?...

[Mathaeus]

Interesting, there is a need for an industry standard for more advanced enveloping methods other than linear blend. Problem is that it can get incredibly complicated very fast so simplicity of use is easily destoyed.

And yes, there's another factor to consider. Export to game engines, where linear blend is only allowed option, almost. So, some kind of automatic creation of intermediate bones, that would be nice one.

Edit : in reality (again imho) one way you can speed up weighting is to create a simple purely volume character for any given rig, e.g. for gear create a fat(ish) character and weight it properly, for each new character you want to weight for that rig, load up fatty and use either its own rig / move the points around by hand with nice big linear falloffs until it matches the current character roughly, gator new mesh to that, clean it up........ not a very elogant solution but could work?...

Yeah, actually I don't know for efficient alternative, when it comes to enveloping of shapes with a lot of small, separated details, like shoes.
By the way, the system I've described, of course is meant to do 100% at once . Just joking, imho small problem with all enveloping is, even the best result looks like a good rubber doll, no more than that. No secondary motion, no sliding around bones... But, nicely distributed points, by machine, could be a good base for additional deformers, imho....

[SpookyMunky]

It could be good yes, but as you admit it's never going to do everything. I have seen promising new ideas, some that can even finally do hands and fingers... I dunno though, the actual weighting part of a character rig I find gets done fast, I'm honestly not sure if a better algorythm would speed up any job since you have to check and check again, make sure its actually worked ?

Iv always found there's a fine line between automation being helpfull and blatantly just taking more time than if you'd done it manually

e.g. even if you have a weighting system that is 98% perfect... you still have to look through it to find the 2% that it will have goosed up, in my mind it could take as long as it would just giving whole mesh 100% weight on a chest bone and working out from there hehe.

Edit: hehe, I should prolly mention I am very ocd though I guess when working .. envelope has to be vertice perfect before I move on to rest of rig.. if I dont have enough time for perfection I fix it with shape deform keys etc

edit2 : also if ya gonna dream, why not dream of opensubd ? ... sigh hehe, its something that's been made, free.. and would change my life completely in a really nice way .. If theres no talks ever of getting opensubd into xsi... as much as it'd suck I might have to move on, lol.. at heart have always been a poly modeller, when lw added metanurbs that was me, perfection.. move to xsi was natural, nicer subd workflow... opensubd to me looks like everything Iv wanted over last 18 years or so ? hehe.. might try creating an xsi version if no-one else does... I suck at that kinda thing tho .. honestly need to go where the best subd is at

[Mathaeus]

e.g. even if you have a weighting system that is 98% perfect... you still have to look through it to find the 2% that it will have goosed up, in my mind it could take as long as it would just giving whole mesh 100% weight on a chest bone and working out from there hehe.

Well I think you know how it is going. The trick is to convince people, that 98% is 'good enough'. 2% is solved by interpolation or another trick, just to avoid the visible error.
Maybe inadequate analogy, maybe not: while ago MR shader called t2s illumination, offered full color fresnel , imho very realistic.For example, gold had transition from yellow on front to light blue on sides.
Later on, MIA comes with monochrome fresnel, but it was built-in thing, also, maybe easier to use. Both were considered as 'realistic', 'correct', what not.
Never underestimate the 'progress' .

[SpookyMunky]

Never underestimate the 'progress' .

You are right , sorry.. just doubt these kinda things(and am depressin scotsman hehe), maybe even 70-90% could be improved to 100% with a few filters and 5mins work.. gueass I should be looking forward to any new improvement in an area that in my mind has gone backwards over the years

[myara]

Yes that what I thought at first... but it didn't worked, Softimage would also start scrambling the already painted weights on other areas by redistributing the smooth influences... this was in 2013, I haven't used Softimage since last year, im now in Maya land, fortunately only as an animator or else I would do crazy

Be sure to freeze your weights frequently, or you'll have some weird bugs all over messing up your weights without an undo option (you'll have to reload your scene).

I recommend to weight in Immediate mode when possible to avoid weight history.

[SpookyMunky]

Dunno if immed mode is necessary ?, I always forget its even an option hehe. get used to the kind of things that will destroy a live stack and how to avoid for the most part. I do save every 5mins to a new file tho just in case hehe, yay for ever increasing hdd sizes ;)

Edit: constant saves to new files are a must btw hehe, has always been my code of practice since I trust nothing, last year on a job for no reason that I could ever figure out at some point while working my scene became poison, for days was working away without restarting xsi hehe, eventually it crashed... every scene going back for days would crash xsi on load attempt hehe, be forewarned ;)

[myara]

Immed is basically an auto-FreezeM function so of course it isn't necessary.

But if you don't have any operator that you want to keep alive under your Modeling stack, then using Immed when weighting or tweaking modeling would keep your data healthy and light without having to freeze manually.

Operators are handy and one of the most powerful features in SI, but if you don't really need them alive and just keep them, they just make your data heavier and SI performance slower. And it could be a little buggy too.