On May 17, TensorFlow launched MoveNet, its latest gesture detection model, as well as a new gesture detection API in TensorFlow.js.
MoveNet is an ultra-fast and accurate model that detects 17 key points in the body. Two variants of this model are available on the TF Hub, namely lightning and Thunder. Lightning is for delay critical applications, while thunder is for applications that require high accuracy. Both models run faster than real-time (30+FPS) on most modern desktops, laptops, and mobile phones, which is proving crucial for real-time fitness, sports, and health applications, and can be achieved by running the model entirely on the client using TensorFlow. There is no need to call the server after the initial page loads, and no dependencies need to be installed.
Can track keypoints through fast motions and withdrawing poses.
Next Generation Pose Detection with MoveNet and TensorFlow.js