A new artificial intelligence system has been created by the researchers from University of California, San Diego, and Adobe Research. The AI is capable of learning people’s fashion styles and create realistic computer-generated images of new fashion items matching their style and likes.
The AI can be divided into two parts. One is a convolutional neural network (CNN) trained to understand the person’s item preferences when fed with purchase data from Amazon in six different categories, including pants, shoes, and tops for both men and women. Such implementations already exist in the form of “You may also like” sections on e-commerce websites.
What’s more surprising is the other half of fashion designer AI which is Generative Adversarial Network (GAN). In the case of GANs, two neural networks are trained with the same set of data. One of them is assigned with the task to create fake images of the data and the other verifies whether the image is real. GANs were first created in 2014 by Ian Goodfellow.
The researchers used the information gathered by the CNN algorithm as the data set for GAN which then created multiple images for each user.
If GAN-based AIs end up on e-commerce websites and online recommendation systems, it could do wonders for them. However, the AI is still in its early stages of development; the researchers are yet to figure out how to convert the system’s 2D renders into 3D which could be used to manufacture clothing.
Also, there is still a long road to go before the AI can create something entirely new. For now, it’s able to throw out some blue shirts for a buyer who likes blue shirts. It would be hard for the AI fashion designer if it were to create a matching pair of shoes to go along with a particular pair of pants.
The potential of AIs in the fashion industry might not be all-capable right now, but still, they could be the next breakthrough in predicting trends. Company’s like Amazon and Alibaba are already working on their versions of fashion designer AIs.
Regarding the AI created by UCSD and Adobe, the Chief Scientist at Vue.ai Costa Colbert said it might be helpful in the case of big names in online retail which requires an extensive amount of data.
You can read the research paper published on ArXiv using this link.