mcgillinnovation / asl-translator

Translates American Sign Language (ASL) to text

SensAI: ASL Translator

Description:
Translates American Sign Language (ASL) to text

Set Up

Set up Python 2.7 environment in Anaconda:
Open Anaconda Prompt and enter the following commands:

conda create --name python27 python=2.7
activate python27
conda install scikit-learn #includes scipy, numpy (?)
conda install theano
pip install keras
pip install opencv-python
conda install cloudpickle
pip install pygame
pip install pyglet
conda install pillow
pip install matplotlib
conda install lasagne
pip install nolearn
pip install h5py
pip install seaborn
pip install imutils

File Structure

conf.json --- configuration file used to provide inputs to entire system; json file with key-value pair file format to store data effectively

• model: takes in any of these parameters - inceptionv3, resnet50, vgg16, vgg19 and xception
• weights: takes the value imagenet specifying that we intend to use weights from imagenet; can also set this None if you wish to train the network from scratch
• include_top: takes the value false specifying that we are going to take the features from any intermediate layer of the network; can set this to true if you want to extract features before the fully connected layers
• test_size: takes the value in the range (0.10 - 0.90), to split overall dataset into training and testing
• seed: takes any value to reproduce same results
• num_classes: specifies the number of classes considered for the image classification problem

Tracker

Key Annotations In Code:

• EDIT! ~ edit
• [?????] or ? ~ question

Commits after training:

• Scenario 1
  Dataset: signs1 (ASL3)
  Pre-trained model: vgg16
  Feature extraction time: 17:26-18:34
  Models/results from test sizes: 0.1, 0.3, 0.5, 0.7, 0.9, 0.95
• Scenario 2
  Dataset: signs1&2 (ASL3, ASL1)
  Pre-trained model: vgg16
  Feature extraction time: 23:34-01:23
  Models/results from test sizes: 0.1, 0.3, 0.5, 0.7, 0.9

Thanks to the following:

Mentors, Sponsors, Support:

• McGill Innovation Team & Participants
• McGill Reasoning & Learning Lab
• CIFAR
• Ethan Macdonald
• Negar Rostamzadeh
• Pedro Oliveira Pinheiro
• Anmol Jawandha
• Perouz Taslakian
• Alexis Smirnov
• Nicolas Le Roux
• Gheorghe Comanici
• Adriana Romero Soriano

Resources:

• Flower Recognition Deep Learning