A modified version of TFJS Pose Detection model (MoveNet.SinglePose.Ligntning) using TFJS React Native in an Expo project. It supports both portrait and landscape mode with front and back camera. Only the keypoints are rendered in the demo.
This project uses Expo SDK 49 and jsc instead of Hermes. FPS is very low on my phone, around 3-4 FPS.
I updated Camera.Constants.Type to CameraType.front/back.
Added 3 additional props to TensorCamera according to this documentation.
Also updated tsconfig.json to use settings for Expo SDK 49.
Update 9/23/2023: Stopping development on this project for now. Will experiment with Flutter and Firebase MLkit.
To run it locally:
$ yarn
$ yarn start
Then scan the QR code to open it in the Expo Go app.
If the app crashes on startup, see here for more info.
