Detecting Numbers Written In Kannada Numerical System using a ResNet-34, a Deep Residual Learning Neural Network for Image Recognition.

  • The DataSet was in a pixel information csv file containing an image of 28x28 pixel.

  • First, I tried to train it using RandomForest Classifier which gave an accuracy score of 97.98 %.

  • Then I switched to ResNet34.

  • I have achieved an accuracy of 99.4 %.

  • To see the project jupyter Notebook click Here

Kannada Numbers Covnverted to image from pixel data using matrix. 28x28

kd

Top Losses after the KannadNet was trained on a 34 layer conv Neural network.

kd

Confusion Matrix

kd

Actual and Predicted images:

kd kd