Convergence: Deep learning enables seeing and interactive virtual trainers
With at-home products commoditizing one-on-one training and computer vision democratizing access to expertise, the stage is set for deep learning, the most recent wave of AI advancement, to combine the two. Imagine the one-on-one connectedness, empowering motivation, and helpful corrective guidance of a personal trainer embodied in an on-demand, non-judgmental, yet human-like avatar. That is TwentyBN’s Millie.
TwentyBN, the AI startup where your editors work, has built
Millie Fit, a computer vision AI-powered fitness coach housed in a life-sized kiosk. Combining video understanding and natural language understanding, Millie offers everything you’d expect from a traditional trainer. Millie leads you through a variety of workouts, offers real-time corrective feedback on your form, and pushes you to hit those last 5 reps. But that is not all what Millie has to offer.
Behind Millie’s capabilities is TwentyBN’s globally distributed data platform that constantly collects high-quality, annotated video data on fitness-related actions from crowdworkers. The data serves as fitness knowledge for the deep learning model inside Millie’s brain. With no need for pose estimation or skeleton mapping over a human body, Millie tracks a wide range of exercises, including those which were previously untrackable by traditional computer vision techniques. Currently, Millie understands over 10 different types of stand-up exercises and over 20 High-Intensity Interval Training (HIIT) exercises. Her expertise in fitness will soon extend to workouts in boxing, yoga, Pilates, and more.