README.md

Classification of American Sign Language

Our goal was to design a classifier that, given an image, is able to understand and recognize the American sign language (ASL) and output the right character represented by the sign.

Table of contents

• Development Setup
• Requirements
  • Download dataset
  • Run
• Best Model
• Demo
• Slides
• Credits

Development Setup

Requirements

conda create --name asl python==3.7.11
conda activate asl
pip install -r requirements.txt

Download Dataset

$ cd dataset
$ python dataset_download.py

Run

python main.py --data-aug  DATA-AUG --target-size TARGET-SIZE --epochs EPOCHS --batch-size BATCH-SIZE --lr LR --name-model NAME-MODEL --fine-tune FINE-TUNE --only-test ONLY-TEST

where:

• DATA-AUG: data aumentation, default is True
• TARGET-SIZE: target size of the image, default is (100, 100)
• EPOCHS: number of epochs, default is 50
• BATCH-SIZE: batch size, default is 64
• LR: learning rate, default is 0.0001
• NAME-MODEL: name of model, defaul is VGG19
• FINE-TUNE: fine tuning, default is False
• ONLY-TEST: only test, default is False

for example, for training

python main.py

for testing:

python main.py --only-test True

Best Model

Demo

ASL.mp4

Slides

Credits

• Marco Pennese

• Sveva Pepe

• Simone Tedeschi