Announcing Valentina: Opensource Fashion Design project

[duststorm]

I think your functional writeup of how this could be implemented is pretty sound. It's similar to what I had in mind.
As of yet not much progress has been made in this project, but this could change in the near future.

[badwolf]

something that may help out is

viewtopic.php?f=9&t=12796

the real trick is going to find what measurement standard is close enough to the MH set to get the mapping setup (im not sure MH has enough numbers to map from)

[bert8]

Wow, new to this community. Have a fashion start-up myself.

Looking for someone who'd be interested in spearheading our 3D initiative. I'm such a newb, when I opened the application I realized I don't know how to manipulate the model at ALL. I feel so out of place, but want to get involved somehow.

[SystemicHatter]

Hi! I actually joined up because I've been fiddling around with Valentina and MakeHuman, and was actually approaching it from the opposite direction: getting more general pattern-making measurements of a customised MakeHuman model for use in Valentina. I'm fairly new to this, so I hope any insights I can provide are helpful.

MakeHuman currently calculates body measurements based on the sum of the straight line distances between a reference set of vertices lying on the mesh (see: getMeasure() method in the modelling measurement extension Python file ). There doesn't necessarily have to be an edge on the mesh between each pair of vertices, so the measurement 'tape' effectively lies on a (hypothetical) triangulated mesh. It also, interestingly, calculates the measurements based on the current pose - so measuring and posing would work for those measurements highly dependent on pose - provided you have the necessary poses to work with. Each measurement also has a target associating it (for mesh deformation).

I guess the question is whether there are vertices on all the spots where 'measuring tape' will be put. If so, could probably mark out the relevant measurements in Blender (assuming mhx or mhx2 preserves the vertex index numbers) and then use them as measurements in MakeHuman. Otherwise, would probably need to interpolate between sets of close-by vertices to place the tape and measure, which gets quickly more complicated.

Measurements which only partially touch the body could be measured by a 'touching the body' part (as above) + getting the height of the vertex from the floor.

Hope that was helpful and not too much of a ramble. Going to play about with trying to mod in some Valentina measurements into MakeHuman.

[duststorm]

Exactly, this is where help from people more familiar with the fashion business are needed, to determine which lines exactly have to be measured.
As all landmarks on the body, which I assume is usually where will be measured (to be able to repeat the same measurement on different bodies I suppose recognizable landmarks are used), have a vertex, so my guess is that we will get by with only marking vertices.

But otherwise something like a measurement proxy could be created, a mesh fitted to the human (just like the clothing) that contains all points that need to be measured.

[badwolf]

in another thread we have a plugin that will allow importing of measurements from a json file (and hopefully we can get a matching export plugin) so now we have the fun of getting the real world measurements mapped to the MH internal values.

[duststorm]

That is easy to do (and can probably be written in an hour).
The hard part is deciding the correct areas to measure, and making the modeler more intelligent (modify multiple modifiers such as macro to get the desired proportions, instead of just scaling naive linear targets -- a complex planning problem that needs a pre-cached cost function to work optimally).

[badwolf]

is there some sort of flowchart type thing that show the order of how the targets are applied??

Im thinking that the age/gender/height targets get us close and then the other targets should get us the rest of the way


(but yes sorting out for measurement X we need to adjust target A B C in that order is going to be the "FUN" part)


i don't suppose that the last core team member to deal with that code can chat??

[duststorm]

is there some sort of flowchart type thing that show the order of how the targets are applied??

No, because there is no order. Or rather, all targets are applied at once, and the calculations involved are perfectly commutative.

i don't suppose that the last core team member to deal with that code can chat??

That would be me.

But indeed this problem can become quite intricate. Especially when there are multiple measurement inputs to be considered.
Finding a best match in that case becomes a numeric optimization problem that has no perfect solution, but needs to be approximated. As you'll be dealing with an overdetermined system, you'll probably have to work with eigen vectors to solve this problem.