ModiFace unveils AI hair transfer tech at CVPR

June 23, 2021

Rendering realistic products on a user’s selfie is what has made ModiFace’s products so successful over the years. Today, we approach this task with scientific rigor, aiming to ensure that the way we mix colors happens in a similar way to how something like a hair colouring product would impact the light that hits your head. But what if we could let artificial intelligence solve this problem for us?

With our method, LOHO, you don't have to imagine --- LOHO uses reverse generative adversarial networks (GANs) to synthesize a photo with your new haircut, color, or both. LOHO is an optimization-based approach that inverts a trained StyleGANv2 model to transfer hair appearance and style from reference hairstyles. Using LOHO for latent space manipulation, users can synthesize new photorealistic images by manipulating hair attributes either individually or jointly. LOHO is a product of a collaboration between Toronto-based industry (ModiFace, a L'Oréal subsidiary) and Vector Institute faculty and students.

Put simply, our algorithm has learned both how to generate an image with our target style and color from constraints it has been trained on. Next, it can take a new image and transfer that style and color to a “rendered” output. It can judge how accurate the results are, so it knows whether or not what it generated was realistic. We give our algorithm an image of the color and style we want to achieve, and the A.I. does the rest! We believe that this will be the future of how we render products, and is a very exciting milestone for our team.