lav30 / kannadanumeralclassifier

Handwritten Numeral Recognition using Convolutional Neural Networks

Kannada Handwritten Digit Recognizer

This project has focused on training a CNN model on handwritten Kannada numerals and also on making predictions in real time using a GUI developed by Gradio. Live predictions are made by drawing digits 0 to 9 in the Kannada language on the drawing pad and the trained model is utilised to make accurate predictions in real time.

Owing to the dataset size and limitations of web deployment , Heroku deployment is currently unavailable. However, local deployment is achieved using Gradio.

Project Description

The dataset can be accessed on Kaggle

Multiclass Classification : High Level View of the Model

Image created using NN-SVG

Results

Several performance metrics have been utilised to determine model generalization capability on the test set. Accuracy, precision and recall have been used to determine model prediction ability. Other metrics such as the ROC curve is also used to determine how well the model generalizes on previously unseen images.

Citation

1. Prabhu, V. (2019, August 03). Kannada-MNIST: A new handwritten DIGITS dataset for the Kannada language. Retrieved April 19, 2021, from https://arxiv.org/abs/1908.01242v1

2. NN-SVG rendering of the model in AlexNet style

License

MIT