I had to google "cosplay market size" to confirm, but the paper's assertion that it's a billion-dollar industry does appear to be correct (!)
Two of the paper's authors work at the technology arm of Zozotown, one of Japan's largest fashion e-commerce sites, which points to how this could have practical benefits.
> I had to google "cosplay market size" to confirm, but the paper's assertion that it's a billion-dollar industry does appear to be correct (!)
It's a thing, I'm taking remote classes now and some seemingly 20 somethings are taking the time to dress up more and more in cosplay attire for them.
I'm all for it as I grew up in the 90s, alongside various Japanese sub-cultures that involved anime/manga, but I admit it is kind of hard to see a lolita with a goth background at 8am lecture for linear algebra or something.
I'd prefer to see it in person, as seeing a cosplay girl online is become blase to me, and would prefer to see if they could pull look off in the summer heat or winter snow.
I frequented a night club before this pandemic where it was typical that dressing for the occasion meant cosplaying to some degree, which meant Halloween was always over-the-top.
Capitalizing on this market is pretty interesting, so many DIYers (some even on hack-a-day inspiration streaks) have become the default for difficult to make things like armour, weapons and helmets. I'm not sure how grass-roots it is anymore, but when I cared to look there were lots of social media posts on sites like Deviant art about their current projects and signups for pre-buys/groupbuys. I'm not sure the West has caught up to Japan's boutique clothing stores that cater to a specific genre of anime quote yet, but I can see there being an incentive now that it is or exceeds being a billion dollar Industry.
If you're interested, this is pretty much the main-plot to a J-movie from the early 2000s called Kamikaze Girls [1], funny movie.
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.
> 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.
A bit offtopic, but my fear with these AI systems is that they will create "Discovery channel" effect in other areas - replace fewer good quality items, with a flood of passable AI generated alternatives, making good quality stuff harder to get. Hope I'm wrong.
From what I've seen this is mostly already the state of the cosplay industry, except the AI part:
* Some big Chinese manufacturers start drafting the easiest way to get a passable-enough version of an outfit to sell on aliexpress for <$50 in time for Halloween /convention season.
* Fans come up with sewing patterns and jewelry sources on their own (a manual process).
This is already happening with targeted clothing on Facebook. The jump from simply scraping FB data and filling in the gaps to AI generating the clothing entirely on its own will probably happen soon.
Interesting enough this is essentially what machine-made lace did for hand-made lace. Superficially a luxury product became mass-market available, which massively reduced the cachet of lace. But machine made lace is also drastically inferior to hand made lace.
This is a cool application, but I was left unimpressed with the final results shown in the paper. It's intriguing that it works at all, yes. But look at the pants and skirts... there is a lot of detail in the original that is lost or obviously distorted in the generated images. Am I expecting too much?
Yeah, I'm failing to see any results here that actually match the original. Sure, they get it impressively close considering the source and the process, but I don't think any of this is actually something that anyone would want to buy.
Two of the paper's authors work at the technology arm of Zozotown, one of Japan's largest fashion e-commerce sites, which points to how this could have practical benefits.