colorcurvature opened this issue on Dec 02, 2010 · 11 posts
colorcurvature posted Thu, 02 December 2010 at 3:32 PM
The discussions about Miki3's problem zones made me think again about the JCMs.
I have used the P8 teaching tool to apply JCM's, but still the possibilities are very limited.
So I was thinking this: Why not pose the body part and make JCMs for the difficult positions where the rigging is not ok.
Record these positions with the proper JCM settings that make a good look.
Then, teach a magic box about the JCMs and in which position they are good.
e.g. bend 0, side-side 0 => no jcm
bend 45, side-side 0 => use jcm_b45_ss_0
bend 45, side_side 45 => use jcm_b45_ss_45
Instead of recording these as ERC parameters (which is a total pain), these are just recorded into the figure as a list.
A python script would read the current pose, and evaluate JCM settings based according to this list. The closer the local body part is to a known good solution (the recorded list), the more influence this solution has.
The various solutions are interpolated accordingly to their distance from the current pose.
This would overcome that linear system, that almost noone can properly program.
Any opinions? It could be programmed with acceptable effort.
colorcurvature posted Fri, 03 December 2010 at 6:39 PM
I have a first version ready. Who wants to help me fine tune it?
Its late here now, but I think I have made a major progress.
I created my first 3 dimensional JCM into the shoulder of P6 James.
And right now with my tired eyes it is looking GREAT!
No matter how I pose this shoulder now, it would choose the best settings for JCMs for me, and this completely free, in all three dimensions at once! No breaks, always smooth!
This might be a step towards the virtually perfect bend. And its working! And its really easily to setup!
Ah, I need help :) Who wants to try it??
colorcurvature posted Fri, 03 December 2010 at 7:04 PM
odf posted Fri, 03 December 2010 at 7:37 PM Online Now!
Looks pretty cool, but I'd probably be more interested in a program that generated the appropriate ERC code for me. It doesn't have to be linear. There's valueOpMultiply and valueOpDivideBy as well. ;)
-- I'm not mad at you, just Westphalian.
colorcurvature posted Sat, 04 December 2010 at 3:23 AM
Ah! I did not know these exist.
But maybe one can generate a classic ERC from it, yes.
With * / and limits, this might be even possible in theory. Will brainstorm a bit today.
Maybe one can create helper parameter for every cell in the grid.
In the 2 dimensional case, with cell grid (i,j), one maybe can create a dial that is ercd in a way that is 1 if the bends are dialed within the dimensions of the grid and 0 otherwise.
Then one can try to solve the problem locally.
What I do here is I have a grid of joint poses and if the figure is posed not on a grid point then there is an interpolation between the corners of the cell the pose is in.
But that is a different problem. Unless one needs animation, I think ERC-by-python is good.
But I will think, thanks for the suggestion
colorcurvature posted Sat, 04 December 2010 at 4:29 AM
Btw, which ERC "patterns" are you using?
Maybe it could start at least with an easy markup for specification (+generation), as I am not sure my grid patterns can be easily translated to legacy erc
amy_aimei posted Sat, 04 December 2010 at 8:52 PM
Hi colorcurvature,
I want to try it and apply it to make conforming clothing for Antonia.
colorcurvature posted Sun, 05 December 2010 at 10:39 AM
i have thought of a model that be generatable and usable with classic ERC
colorcurvature posted Sun, 05 December 2010 at 10:46 AM
with valueOpMult, I think one can model the function abs(a):
abs(a) = min_limit_0( a ) valueOpAdd min_limit_0( a valueOpMult -1 )
with abs(a), one can define the influence of a 2 dimensional jcm
like min_limit_0( 1 - abs(distance_x)+abs(distance_y) )
this leads to the square shaped influence pattern
the 2d space is overlappable by squares like shown above, so that the sum of all influencing JCM's is 1 at each point in space (interpolation).
if i can verify this and prototype it into a figure, then this would mean that we indeed can generate ERCs for Poser Legacy ERC
R_Hatch posted Sun, 05 December 2010 at 11:24 PM
Sorry I haven't replied earlier, Hannes.
I must have been tired, because I saw the title and somehow got totally the wrong idea, when in fact, I am currently facing exactly this issue. Once you get something ready, I'll take a look and run it through its paces :)
colorcurvature posted Mon, 06 December 2010 at 1:58 PM
You are welcome :)
I uploaded usage instructions here: cgscripts.colorcurvature.com/EasyJCM.pdf
The prototype will come by e-mail