Computer Vision: Sign Language Digits Recognition

The kaggle task: https://www.kaggle.com/javaidahmadwani/sign-language-digits-dataset

Dataset

This data set contains images of sign language digits. There are ten classes, labeled as 0 through 9, and each class is made up of images of hands showing the sign for that particular digit.

Each class has between 204 and 208 samples. The total data set contains 2062 samples.

• Image size: 100 x 100 pixels
• Color space: RGB
• Number of classes: 10 (Digits: 0-9)

Libraries

pip install -r requirements.txt

• numPy
• tensorFlow
• pickle
• PIL
• glob

How to run?

To test the model on your own photo you just have to run the main.py file. You'll be able to upload your photo(doesn't handle .heic ext).

             

Neural Network Models

Current best accuracy on the cross-validation set: 95.7%

Model	Train Set Accuracy	CV Set Accuracy
Custom convNet	99.8%	95.7%
Resnet	97.4%	89.3%

Custom ConvNet Structure

Confusion Matrix (Custom convNet)