Skip to content

A sign language image classification using transfer learning from MobileNet model.

License

Notifications You must be signed in to change notification settings

melihaltun/sign_language_image_recognition

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 

Repository files navigation

SignLanguageImageRecognition

A sign language image classification using transfer learning from MobileNet model.

A deep CNN implementation that modifies the last few layers of the pretrained MobileNet model and repurposes it to classify sign language digit images.

Depending on the selected subset of training and testing images, 99 to 100% accuracy is usually achieved.

The code needs tensorflow, keras, numpy, pandas, matplotlib, sklearn, itertools and glob installed in the Python environment.

Download the dataset from: https://github.com/ardamavi/Sign-Language-Digits-Dataset image

Extract the contents of train folder to "./Sign-Language-Digits-Dataset/" folder under the project folder

During the first run the code will do a random sampling of the data and form the train, test, validation sets. They will be reused in the subsequent runs. Number of images in each set can be modified as needed. If a new train, validation & test set is needed, delete the "test", "train", and "valid" folders under "./Sign-Language-Digits-Dataset/"

The code will save the trained model and use it the next time it is called. To retrain a model, go to "./models/" folder under the project library and delete the "sign_language_model.h5" file.

GPU parallelization is turned off, but it can be turned on by uncommenting the relevant line.

For GPU utilization, a combination of Python 3.8, Tensorflow 2.10.0, CUDA 11.2, CUDNN 8.8.1 and Zlib works for this model, although other configurations may also be possible.

About

A sign language image classification using transfer learning from MobileNet model.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages