Hans Pose action classifier

Current action classifier takes human body topology into account, and thus is designed for human body pose only. For hand pose action classification, 1) if your use case is simple, you may not need an ML model, just read out the hand pose coordinates, and define some rules or heuristics for your cases. 2) if your case involves more complicated hand actions, you may experiment on last year's model: https://developer.apple.com/videos/play/wwdc2019/426/. This model is more generic and also time-series based, but requires you to extract the hand pose points, prepare them into CSV files, etc. and train. 3) if you use other ML tools, you also need to extract hand pose points to files, train a model, and then convert it to CoreML, and run on Apple devices. 4) stay tuned!

Current body action classifier is not suggested for hand. However, if your use cases are not too complicated, you may define some rules or heuristics in your App logic, by checking the hand pose joints coordinates, for example.