After clicking “Watch Now” you will be prompted to login or join.
Google Cloud AutoML Video and Edge Deployment
Yongzhe Wang, Google
GTC 2020
Google Cloud will be presenting AutoML Edge Video, a solution using NVIDIA GPUs. AutoML Edge Video allows users to train customized models without knowing how to tune parameters, using Google's AutoML. In this talk we will focus on our end to end solution for Video Classification and Video Object tracking. To train a model, we use multiple NVIDIA GPUs in parallel for hyper-parameter tuning and transfer learning, allowing us to return high-performance models quickly. We will show how to export models to the edge using a frozen Tensorflow graph, and how to optimize for NVIDIA GPUs including Jetson and Tesla T4.