This repository has no license, therefore all rights are reserved. You cannot modily or redistribute this code without explicit permission from the copyright holder.
This is a specific use-case of the Real-time Action Recognition System. In this implementation, we use the new model to demostrate handwash auditing, by using a live feed of a camera setup above a sink where the handwash is being performed.
a. In the parent directory, run:
$ pip install -r requirements.txt
a. Change directory to pyflow
$ cd pyflow
b. Setup pyflow
:
$ python setup.py build_ext -i
a. Download the model from here.
b. Save the model in the parent folder of this repository, with the name: current_final_handwash_model.h5
$ python app.py
5. Open localhost:8001
on your browser when the * Debugger pin is active
message appears on the Terminal.
The training and testing data used was the first split of the UCF-101 Dataset. Due to computation limitations, the first split of the UCF-101 dataset has been preprocessed accoring to our implementation, and stored in the following locations as .pyc
files:
The pre-processed training data can be found here.
The pre-processed testing data can be found here.
A Kaggle dataset for this application can be found here.
- A webcam must be connected to the computer/laptop.
- The model must be saved in this repository's parent folder.
- The computer must have internet access, since JavaScript is dynamically linked.