Vocalization sign language with deep learning.
Vocalization sign language with deep learning.
In this project we use our own Sign Language Dataset
.
Vocalization sign language iOS App: Vocalization-Sign-Language-iOS
Demo |
---|
Watch Demo Videos 🔊 |
In this project, I added deep learning to my old lip reading project SesimVar(Turkish).
Note: If you are failed, look up For Development
title in bellow.
python3 live.py
Note: If you want, you can change the delay time.
python3 predict.py <ImageFileName>
cd Data && chmod +x download_dataset.sh && ./download_dataset.sh
(Updated) Model
coming soon!Demo |
---|
Watch Demo Videos 🔊 |
DataSet Examples | Model Accuracy |
---|---|
In this project we use our own Sign Language Digits Dataset.
At the end of 25 epochs, 97% accuracy was achieved in the test without data augmentation:
Input Data Shape: 64x64x1
Convolutional Layer 32 filter Filter shape: 3x3 Strides: 1x1 Padding: Same
Activation Function: ReLu
Convolutional Layer 64 filter Filter shape: 3x3 Strides: 1x1 Padding: Same
Activation Function: ReLu
Max Pooling Pool shape: 2x2 Strides: 2x2
Convolutional Layer 64 filter Filter shape: 3x3 Strides: 1x1 Padding: Same
Activation Function: ReLu
Max Pooling Pool shape: 2x2 Strides: 2x2
Convolutional Layer 128 filter Filter shape: 3x3 Strides: 1x1 Padding: Same
Activation Function: ReLu
Max Pooling Pool shape: 2x2 Strides: 2x2
Flatten
Dense Size: 526
Activation Function: ReLu
Dropout Rate: 0.5
Dense Size: 128
Activation Function: ReLu
Dropout Rate: 0.5
Dense Size: Class size in dataset
Activation Function: Softmax
Total params: 4,507,864
Trainable params: 4,507,864
Non-trainable params: 0
python3 train.py
Not forget to download dataset before training!
tensorboard --logdir=Data/Checkpoints/logs
For getting dataset look up Getting Dataset title in this file.
For your own dataset:
sudo pip3 install -r requirements.txt
command.Mavi, A., (2020), “A New Dataset and Proposed Convolutional Neural Network Architecture for Classification of American Sign Language Digits”, arXiv:2011.08927 [cs.CV]