This project uses computer vision to track hand motion and machine learning to interpret the intended effect of the hand motion, in terms of tempo, dynamics, and attack.
this program doesn't necessarily need to installed, but all dependencies necessary to run the python script need to be met. This program is written in python 2.7.
requirements.txt
has all of the python dependencies. Install them with
pip install -r requirements.txt
To track hand movement, this project makes use of tensorflow models' Object Detection API. Follow this link to install the object_detection api, and make sure that the research and slim are in the PYTHONPATH while running this program.
python create_dataset.py
creates tfrecord dataset used to train hand tracking model. Not necessary unless used to extend or improve the hand tracking model. NOTE: this script creates approximately 450 MB of new data.
python model_main.py --logtostderr --model_dir=./model/ --pipeline_config_path=model/pipeline.config
re-trains the hand detection model. NOTE: must have TFRecord files created by running python create_dataset.py
python conductAR.py
takes the included pre-trained model to track hand motion.