An algorithm that automatically fits clothing?

Question

Imagine a client asked you to design a piece of software that would satisfy some pretty rough specs as follows:

1: it will be marketed towards the fashion industry

2: the users will be people that "design clothes and stuff" (there is likely a specific term for it but it doesn't come to mind)

3: For various reasons, it would be useful to be able to quickly prototype designs and see how they look on models. To make it more useful, it should be rendering 3D models so that the designers can look at it from various angles if necessary.

4: In order to be useful, users should be able to design their own outfits using whatever 3D modeling tool they like (it may even be built-in), and then import it into this system. Similarly, users should also be able to design their own models as well.

5: Users will be able to pick a model that they wish to try the outfit on. Naturally, humans come in various proportions so they may want to adjust the model accordingly to see how it looks on different people.

6: Consequently, the outfits would need to be adjusted to account for the change in proportions. It is not feasible to have to design the same outfit over and over again just to make some minor changes, so this should be done by the computer.

Most of the system sounds pretty straightforward. But then comes the 3D stuff at the end.

In particular, the most difficult part of the system I imagine would be the actual "clothing fitting" algorithms, since you can create static objects and see how they look, but to be able to change them on the fly and still have them fit naturally seems absolutely mind-blowing.

Are there any tools like this already? What kind of tools would make this easier to solve? Has anyone ever worked on something like this?

Answer 1

There's a lot of hardcore computational geometry in this, and while bits and pieces of it do exist, as far as I know no one's done the whole thing before.

Oddly enough, the clothing fitting may actually be one of the easier parts. You would have the designer define a fabric pattern with parametrically sized components, simulated a model wearing it at a particular size and in various poses, and produce a measure of tightness (amount of deformation) in each size-pose pair. Then you'd define a cost function with a penalty for size and for tightness, then do gradient descent to minimize it.

But first you need fabric simulation (hard), 3D modeling of humans in various poses (hard), and a modeling tool for parametrically shaped fabric patterns (hard).