New App Turns Your Selfie Into a Personalized Emoji

San Francisco-based startup Mirror AI released the beta version of their deep learning-based mobile app that instantly turns your selfie into a collection of custom emojis that look like you.

“[Mirror Emoji Keyboard] is the culmination of a year’s worth of work from our dedicated and experienced engineering team and is only the beginning,” Serge Faguet, CEO and co-founder of Mirror AI. “As our user base grows, our technology will continue to improve, leading to even greater personalization as well as automatic detection of clothing, pets and frequent companions in your camera roll.”

The startup has raised $3.5 million from investors including NBA star Kevin Durant, SoftBank Group, Greylock Ventures, SV Angel, Peter Thiel, Y Combinator, Peter Diamandis, and others.

Using Tesla GPUs on Amazon Web Services with the cuDNN-accelerated Caffe deep learning framework, the team trained their neural network to identify unique facial characteristics from your selfie and then turns those features into an emoji in seconds.

Faguet mentioned the inference happens on Amazon GPU-based cloud instances since they use a lot of state-of-the-art methods and mobile hardware just isn’t ready for that yet.

The app also allows you to create a custom emoji with others – simply snap a photo of your friend and the app will display group emojis.

“An entire section of our brain is dedicated to human facial perception which is an incredibly hard problem for artificial intelligence approaches to solve and we are just starting to scratch the surface of the possibilities,” Mirror’s CTO and co-founder, Evgeny Kuryshev, said. “We are fortunate to have some of the top engineers in this field, which allows us to relentlessly work to improve this technology. A decade from now, Mirror aims to deploy our technology and usage data for complex tasks from powering virtual reality characters to advertising personalized to viewers’ faces.”

The Mirror Emoji Keyboard app is available for iOS and Android users.

Read more >