Would you wear a dress with a large hole in it? How about a pair of trousers with two extra legs? Researchers at Facebook have used artificially intelligent algorithms to design a 1,000-piece collection of grunge-inspired tees, dresses, jumpers, and more – and let’s just say it’s not very good.
The resulting designs, published on arxiv, might not be ideal as far as practicality and style are concerned (see above), but the team hopes the computer-made clothing will provide a flash of inspiration for living, breathing, human designers.
The process involves something called a Generative Adversarial Network (GAN), which pits two neural networks trained on a large database of fashion images against one another. The first network comes up with the idea. The second network then approves the design or shoots it down. Over time, the pair hone their skills, theoretically at least, becoming better and better designers.
The researchers added in a third network (called StyleGAN) to limit the designs to wearable shapes.
One problem with GANs is that they lack originality – a trait many an artist prides themselves on. Instead, the designs they generate tend to look suspiciously similar to the clothes they were exposed to during training. So, to try and spur on a little more creativity, the researchers built the networks with two disruptive functions, which confuse the networks and cause them to deviate from "the norm".
While the GANs don’t seem to pose much of a threat to the Karl Lagerfelds and Stella McCartneys of the world, 800 randomly selected items were able to successfully convince a panel of people the designs were human-made 61 percent of the time.
This isn't the first artificially intelligent program to attempt fashion design, and Facebook isn't even the first tech giant to jump on this particular bandwagon. Last summer, Amazon used GANs to produce simple items of clothing based on a particular fashion style they had been exposed to. In the not-so-distant future, the technology could be used to generate custom fashion for consumers based on their style preferences or past purchases.
“Technology can amplify our creativity,” said Camille Couprie, a research scientist at Facebook Artificial Intelligence Research, told New Scientist. “It can take a lot of brain power to think about new ideas, so if AI can help with that and accelerate that process it would be good.”
But for now, at least, it appears fashion designers are safe from the coming of the job-snatching robots.
[H/T: New Scientist]