After clicking “Watch Now” you will be prompted to login or join.
Implementing AI-Powered Semantic Character Recognition in Motor Racing Sports
David D. Albarracín Molina, Virtually Live | Jesús Hormigo, Virtually Live
GTC 2020
We'll present a use case of deep learning for computer vision applied to live TV production of racing events. Currently, the system can identify, point, and label racers on motor sports footage in a fully autonomous way. Its first real-world application has been in live re-transmissions of Formula E events. We'll explain the challenges found and solved in the path from polishing a business idea to delivering a working product to the client, Formula E TV, in the first race of the season. We'll talk about using Tensorflow to build a state-of-the-art neural network and the particular training pipeline designed for the case. We'll also describe our optimized inferencer, ready for the following season, which uses TensorRT to enable the application to run at 25fps in a Jetson Xavier, and over 4x faster on a GeForce RTX2080Ti. We'll close with a 10 minute Q&A about the technologies and the product itself.