Parametric modeling?

[Artist]

MakeHuman has symmetry buttons in the top toolbar for precisely this.

The symmetry button works where there are separate left and right objects, such as ears and eyes. There are other controls that produce asymmetries that this button has no effect on. Two of them are:

Modeling > Torso > Hip >Trans (front view left to right)
Modeling > Torso > Hip >Trans (side view back to forward)

I changed these before I realized what they did and found I could not tell when the slider was exactly centered to get the precise symmetry needed. So I had to open the mhm file in a word processor in an attempt to identify these settings in there and set them to zero. I could not identify them with total confidence. So I had to create mhm files that had only one slider position different from the other. Upon comparison I found that the first of the above sliders had the parameter name "torso hip-trans-in-out" (not an intuitive name) and the second one "Hip-trans-forward-backward". With this known I used the word processor to set these two values to zero and regain the needed symmetry. This took some time to do, and I feel something is wrong with a GUI if something such as this has to be done. Had I been able to enter zero in the GUI I would not have had to.

[duststorm]

There are other controls that produce asymmetries that this button has no effect on. Two of them are:

Modeling > Torso > Hip >Trans (front view left to right)
Modeling > Torso > Hip >Trans (side view back to forward)

You can right click sliders to reset them to their default position. That will center these sliders.

"torso hip-trans-in-out" (not an intuitive name)

I know, I've complained about these at length too...

[zoiner]

Thank you for explanation. If the system is overdetermined it seems as if some form of steepest slope interation is necessary but the problem may be multiple best fit minima with constraints. That is overkill.

From my viewpoint the age, gender, height and weight are usually available so a process of iterating to the height and then weight on a standard model will probably be close enough.
Occasionally I might to have to model non-standard segment ratios (eg hip to thigh vs. knee to ankle) but this would be very rare. So as it stands I will go with the procedure described.

Just a thought. As we tweak the sliders, the program calculates the various values anyway. Rather than have different numerical data entry and storage, how hard would it be to have something like an optional "goalseek" box for a slider. Enable the box, enter the target, then let the program just iterate up or down a bit until within some tolerance. That way the entire numerical storage and internal representation stays in the same units. The calculation procedure is unaltered. Nothing else has to change.

When I click on a slider currently the program takes information from my click and interprets that as a value. All you need is a routine that imitates mouse clicks. 1) Enter a Target 2) simply auto generate say 4 mouse clicks along the slider, generate four results pick the interval containing the target, repeat the process with smaller steps etc and then hand back control when close enough.

For those who would want this facility the repeated calculation time is nothing in comparson with saving multiple figures, noting the extremes, and interpolating, while for everyone else there is no loss of functionality.

I am very busy at the moment and my knowledge of Python is cr*p as far as GUI goes {I just make Python imitate my old pascal programs in console mode. What's an object??? as for classes..!!!} so I doubt if I can help. I bet the source code will be completely incomprehensible to me but I will take a look when I get a chance.

[duststorm]

how hard would it be to have something like an optional "goalseek" box for a slider.

Algo is already implemented in measurement plugin

All you need is a routine that imitates mouse clicks.

You don't need to simulate clicks to update modifiers.

[zoiner]

Thank you!!

I didn't try the measurement plugin when I fired up v1.0.0. It does everything except stature as far as I can see. It will do all I will ever need and more.

In passing why are chest, waist and hips statistics shown to 9 significant figures behind the decimal place?

In reality measurement to 1cm is good unless using a real anthropometer. Measurement to 1mm is exceptional and adequate for any further significant calculations.

I personally would consider cited accuracy to 1mm brilliant and as long as I were within 0.5cm I would be pleased. (I remember as a physics undergraduate being expected to measure object dimensions within 0.1mm using a metre rule that had been calibrated against a standard; but achieving such accuracy with living, compressible, and moving objects is impossible.)

It is sensible to carry all significant figure within the calculations to minimise rounding errors. When the final figures are shown, producing a value to 0.1 mm will indicate whether there is any potential for rounding error at 1mm. Going beyond that will be confusing.

Thanks again.

[marcocap]

Hi All,

This thread is particularly interesting for random or semi-random generation of morphologies.

I am now trying to build a set of morphologies starting from circumferences.

like Burst 90 cm, Waist 70 cm, Hips 90 cm => 5 different morphologies, all other Degrees of Freedom left free to change.

Does Anyone have some feedbacks concerning this neural networks approach ? or any other that could help ?

Thanks all,

Marc