- Initial usage of this project was used for the detection of the various signs in the ASL and to help our end goal of integrating this in a mobile app to help the mute people better communicate with people who aren't cognizant of the language.
-
This project structure additionally helps to accelerate the development of image recognition models in a quick manner.
-
All you have to know is that the preprocessing code is present in the
prepare_data.py
file if you are not bothered about the colour of the training images and only want to focus on the training then don't change the code here
-
Use the
train_model.py
file to train your model. But first you should modify the path parameters as per the dataset you are using. -
Place your dataset in the dataset folder. Then modify the parameters section in the
train_model.py
-
The various parameters that you would want to set would be
num_classes = Number of classes you want to classify eg. 29
>no_of_epochs = Required number of epochs or iterations you want. eg. 10
>size = A list of the width and height of the image for eg. [200, 200]
>batch_size = Here you define the number of images to be fed. eg. 32
- More adjustments can be done to the section of:-
Network Parameters
>Optimization Parameters
-
Here you will define the Network architecture that you want.
-
Based on the network that you have defined remember to make changes to the network parameters in the
train_model.py
file.
-
This file is still under development as it is the file that will carry out the detection part (the final stage) of the whole project.
-
It contains hardcoded list of labels and preprocessing which will soon be converted to reading from
csv
and making a separate funciton for preprocessing the image(frame) captured from the video camera.