4 ms·

For some reason I expected this to be from anime image to costume sewing pattern.

Hextinium 6y ago
Same, possibly by taking 3d models and flattening them or something. Paper is interesting none the less.
Animats 6y ago
Right. What it really does is take an image of a character in an arbitrary, but mostly standing, position, and returns an image in a standard resting stand position. This is neat, but costume generation is several steps past that. Now, if you could take a collection of pictures of a character from anime and get a 3D model you could feed a fashion CAD system, that would be impressive. There is CAD for fashion, and it's very good.[1] The CAD systems understand cloth physics quite well now. A machine learning system which takes a set of pictures of a character in different poses and tries to make clothing which matches would be a useful achievement. It's a big job, but probably do-able. The problem is reasonably convex. If you have a character in a dress, and you start with a generic dress in the CAD system, you can keep tweaking parameters and textures until it matches. The ML system doesn't have to invent a 3D dress, just keep tweaking a basic form. [1] https://clo3d.com/ https://clo3d.com/
- aidenn0 6y ago
  This probably takes us 1 step closer, right? Normalization often a first step to recognition.
- bsder 6y ago
  > The CAD systems understand cloth physics quite well now. Is this actually true? Energy distribution in draping cloth is a stiff differential equation system with multiple time constants--a very difficult system to simulate. We have some decent heuristics that kinda make things work. (The pictures from the link you provided look like those heuristics to my eye). And in fashion they may actually be "good enough". But I haven't seen a genuine advance in computer science/graphics for simulating realistic cloth other than "stupid amounts of compute". I'm willing to be corrected if somebody has references. Edit: Looking at this video: https://www.youtube.com/watch?v=ly7UzwB6MbU&feature=youtu.be https://www.youtube.com/watch?v=ly7UzwB6MbU&feature=youtu.be Heuristics with low-density point cloud are in operation--which, as I point out, may actually be fine for fashion.

Same, possibly by taking 3d models and flattening them or something. Paper is interesting none the less.

Right. What it really does is take an image of a character in an arbitrary, but mostly standing, position, and returns an image in a standard resting stand position. This is neat, but costume generation is several steps past that. Now, if you could take a collection of pictures of a character from anime and get a 3D model you could feed a fashion CAD system, that would be impressive. There is CAD for fashion, and it's very good.[1] The CAD systems understand cloth physics quite well now. A machine learning system which takes a set of pictures of a character in different poses and tries to make clothing which matches would be a useful achievement. It's a big job, but probably do-able. The problem is reasonably convex. If you have a character in a dress, and you start with a generic dress in the CAD system, you can keep tweaking parameters and textures until it matches. The ML system doesn't have to invent a 3D dress, just keep tweaking a basic form. [1] https://clo3d.com/ https://clo3d.com/

aidenn0 6y ago
This probably takes us 1 step closer, right? Normalization often a first step to recognition.
bsder 6y ago
> The CAD systems understand cloth physics quite well now. Is this actually true? Energy distribution in draping cloth is a stiff differential equation system with multiple time constants--a very difficult system to simulate. We have some decent heuristics that kinda make things work. (The pictures from the link you provided look like those heuristics to my eye). And in fashion they may actually be "good enough". But I haven't seen a genuine advance in computer science/graphics for simulating realistic cloth other than "stupid amounts of compute". I'm willing to be corrected if somebody has references. Edit: Looking at this video: https://www.youtube.com/watch?v=ly7UzwB6MbU&feature=youtu.be https://www.youtube.com/watch?v=ly7UzwB6MbU&feature=youtu.be Heuristics with low-density point cloud are in operation--which, as I point out, may actually be fine for fashion.