Productivity Tracker is a simple website to help you stay productive in front of a computer. It utilizes machine learning to make real-time productivity level prediction based on web cam input.
Install NodeJS and Python in your system.
./backend: Code to train a model offline.
./static_website: Code for a simple website demo.
./data : Directory to download datasets.
./generate_labels.py: Python script to generate labels from datasets.
We define productivty by labelling some certain range in valence and arousal as non-productive plus manual annotation on top. Therefore we utlize public datasets with existing valence and arousal annotations. See readmes in data folder for detail.
To generate non-productive frames, run python generate_labels.py.
To make use of existing tfjs models (MobileNet and face detector BlazeFace. We use tfjs in NodeJS to extract feature from datasets and save them as npy. Then we train a model with regular Keras by taking saved npy as input.
cd backend
npm install
node generate_npy.js
cd backend + python train.py
Convert to tfjs model:
tensorflowjs_converter --input_format=tf_saved_model --output_format tfjs_graph_model backend/saved_models/best_model /saved_models/model_tfjs
Static Website is located at static_website, deployed in Productivity Tracker.
Nashila Jahan