Beauty may only be skin deep, but that hasn’t slowed a multi-billion dollar skincare industry.
Now researchers using deep learning are bringing an unblinking eye to wrinkles, age spots and other biomarkers that can signal the state of health of a person’s skin — and helping determine the effectiveness of various treatments.
“How do you evaluate your skin condition, when most feedback is biased?” asked data scientist Konstantin Kiselev during his presentation at the GPU Technology Conference. His firm, Baltimore, Maryland-based Youth Laboratories, aims to better test the results of skin treatments and suggest ways to improve the condition of the body’s largest organ.
His research team’s early work involved preparing 300 manually tagged images for wrinkles. They also ran tests using a VGG network and SegNet, a deep encoder-decoder architecture for pixel segmentation. Then, deploying an NVIDIA Tesla K80 for training and testing provided a 20x speedup over a CPU, Kiselev said.
The team then created an facial detection app, called RYNKL, to create a “wrinkles map zone” covering the forehead, eyes, cheeks and mouth. The app processes each area and assigns a number, which, once added up, provides a RYNKL score.
To popularize the app (available in beta on Google Play and Apple’s AppStore), the team developed a platform for testing various algorithms that rate human attractiveness.
For a novel twist, earlier this year they launched Beauty.AI, the first online beauty pageant judged by an all-robot jury. Users could submit pictures linked to their age through the website. A second pageant is set for May.
Technical advancements in computer vision, facial recognition, machine learning, and neural networks have allowed Kiselev and team to determine facial skin biomarkers using a photo of someone’s face.
Ultimately, they hope to develop a set of user-friendly applications for other aging biomarkers, the ability to detect skin diseases, and make personalized recommendations for skin treatments.