EE541 Final Project - Spring 2022 - American Sign Language Reader using Deep Learning Algorithms

University of Southern California

Contributors

Rajath Ramegowda
Rudrax Dave

Note : Summarised Report is added in the current repository

---

Introduction

This project allows you to put your abilities learned in Introduction to Deep Learning in Python and Image Processing with Pytorch in Python into practice, such as creating convolutional neural networks to categorize photos. Computer vision systems that convert sign language to spoken language have made significant development in recent years. Complex neural network topologies are frequently used in this technology to identify tiny patterns in live video. However, understanding how to construct a translation system is the first step. In this notebook, we'll learn to categorize photos of American Sign Language (ASL) letters using a convolutional neural network. We will train the network and assess its performance after loading, inspecting, and preparing the data.

Goal

The goal of this project is to build a deep learning model to read the American Sign Language. Building 6 CNN models and comparing performance with predefined ResNet50 model.

Results

CNN 3 convolution layer with 89.99% accuracy ResNet50 with 100% accuracy

---

Note to access project files

CNN Models ipynb files

cnn_1Layer
cnn_3Layer
cnn_5Layer

ResNet50 ipynb file

ResNet50

Report