Skip to content

Latest commit

 

History

History
36 lines (24 loc) · 1.88 KB

README.md

File metadata and controls

36 lines (24 loc) · 1.88 KB

tensorglove

A hand pose recognition library using Tensorflow, for use with the Noitom Hi5 gloves. This project was the result of 2 hackathons, so is fairly preliminary.

Alt Text

Getting Started

The repository consists of two components: a Unity project for performing the hand tracking in VR, and a set of python scripts for performing the machine learning.

To train the neural network and run the server, the following packages are required:

  • Tensorflow 1.7
  • python-osc (for running the server).
  • pandas

Simply run the script training.py to train the neural network. After training, you can run predict_server.py to start serving predictions over OSC.

To run the Unity project, the following plugins will need to be added to the project (in the glovetest/Assets folder):

The Standard Character Assets must also be imported, which can be done in Unity from Assets/Import Package/Characters

Open the Unity project (root at the folder glovetest) and open the GloveGestureTest scene.

The Hi5CsvRecorder script in the hierarchy can be used to record poses for use in training the neural network. Select which gesture you are training for (currently from None, Click, Fist and Point) and press L or R to record data for the left or right hand respectively.

The TensorflowOscGesture script can be used to communicate with the tensorglove server. The basic demo only works for the right hand and changes the colour of the hand based on the gesture detected.