FingersCounter

CNN multiclass classifier for counting number of fingers shown on one hand. Made with EduNet.

By Matas Gumbinas, 2019

Introduction & acknowledgements

The sole goal of this work is to demonstrate EduNet neural networks modeling framework [1].
For this task a Kaggle dataset "Fingers" by Pavel Koryakin [2] was used. The purpose of this
dataset is to help create an image classifier that would count fingers on a hand from 0 to 5.

Results:
During this work, using EduNet API, a Convolutional Neural Network was created to be trained on the
given dataset. Afterwards, the trained model was tested using unused part of the dataset meant for
testing the trained models. The trained model reached 100% accuracy with training, validation and
testing datasets after 4 epochs.

List of references:

1. EduNet. Numpy based educational neural networks modeling framework - from scratch
   (by Matas Gumbinas).
   https://github.com/gMatas/edunet
2. "Fingers" dataset (by Pavel Koryakin).
   https://www.kaggle.com/koryakinp/fingers

Starting up

First, download EduNet python package from github [1] (v1.3.0-alpha.0 release) to the main
directory (at fingers-counter/). Or run the following git commands from your console/terminal
starting from the main directory:

git clone https://github.com/gmatas/edunet
cd edunet
git checkout v1.3.0-alpha.0
cd ..

Next, from the fingers-counter/project main directory run the following pip command to install the
required python dependencies:

pip install -r requirements.txt

Finaly, before running this notebook, please run following commands:

cd datasets
python fingers.py
cd ..

fingers.py script will unzip "fingers.zip" file and read its contents to original "fingers"
dataset. Afterwards it will generate a downsampled version of it to a directory "fingers_32x32".
The following work will be done on the downsampled dataset.

That's it! Have fun using this project!