University of California researchers and Adobe have come up with a way for artificial intelligence to understand a person’s fashion style and then come up with images of items that match it.
And, in a recently published paper, they say they believe the platform could be a first step towards building systems that go beyond recommending existing items from clothing retailer’s lines and can be used to suggest styles and aid the design of brand new products.
The researchers point out that recommendation systems have to date had problems when it comes to fashion as the library of products is long-tailed while new ones are continually introduced. Also, users’ preferences and product styles change over time, while the semantics that determine what is ‘fashionable’ are incredibly complex.
The breakthrough here is adding generative adversarial networks (GANs) – AI able to generate realistic images – to the mix. GANs came to prominence recently when the neural networks were able to create pretty credible fake celebrity faces.
“We believe this opens up a promising line of work in using recommender systems for design. Other than improving the quality of the generated images and providing control of fine-grained styles, the same ideas can be applied to visual data besides fashion images, or even to non-visual forms of content,” the researchers say.
Image: Steve Mullins