To learn about what LiveSigns is, look at our devposts description. Watch our video demo!
Firstly, clone this repository and checkout to the web branch. Run
pip install -r requirements.txt
or manually install dependencies as necessary, and run
streamlit run web2.py
or
streamlit run web.py
for the original website version.
This should run the website on your local computer, and streamlit will give you instructions on how to access it.
Checkout to the main branch. Run
pip install -r requirements.txt
or manually install dependencies as necessary, and run
python3 asl_camera.py
This will open up OBS virtual camera with LiveSigns working. You can now go onto any video conference platform and turn on your camera (select "OBS Virtual Camera" for your camera).
You must have OBS and the OBS virtual camera add-on installed.
Images used to generate the dataset come from Akash's ASL Alphabet
Executing python database.py will then genereate a training.csv file.
Using a maxium of 2800 training images per letter/symbol, it was possible to generate this dataset.
With the previously generated dataset, 6 training algorithms (Logistic Regression, Decision Tree Classifier, Random Forest Classifier, Gaussian NB, Linear Discriminant Analysis) were tried in order to obtain the highest accuracy from the testing set.
================================================
EVALUATING MODEL RANDOM FOREST CLASSIFIER
================================================
precision recall f1-score support
a 0.97 0.97 0.97 95
b 0.98 1.00 0.99 113
c 0.95 1.00 0.97 97
d 0.97 0.93 0.95 100
del 0.95 0.96 0.96 106
e 1.00 0.99 0.99 99
f 0.99 0.98 0.98 96
g 1.00 1.00 1.00 95
h 1.00 1.00 1.00 103
i 0.98 1.00 0.99 90
j 1.00 0.97 0.98 100
k 0.98 1.00 0.99 93
l 0.98 0.96 0.97 113
m 0.86 0.98 0.92 98
n 0.95 0.91 0.93 104
o 0.96 0.95 0.96 103
p 1.00 0.98 0.99 100
q 0.97 0.99 0.98 94
r 0.99 0.98 0.99 110
s 0.97 0.95 0.96 93
space 0.98 0.97 0.98 111
t 0.99 0.93 0.96 101
u 0.95 0.93 0.94 85
v 0.95 0.98 0.97 106
w 0.99 1.00 1.00 109
x 0.97 0.95 0.96 92
y 0.97 0.98 0.97 87
z 1.00 1.00 1.00 107
accuracy 0.97 2800
macro avg 0.97 0.97 0.97 2800
weighted avg 0.97 0.97 0.97 2800