Bangla-Handwritten-Character-Recognition

Bangla Handwritten Character Recognition using different CNN architectures.

Table of Contents

1. Overview
2. Keras
3. Pytorch
4. License
5. Contact

Overview

The handwritten character recognition problem which varies among different languages due to distinct shapes, strokes and number of characters has been studied extensively during the last few decades with varying level of success. In case of Bangla Handwritten Character Recognition, here I tried different CNN architectures and made few modifications to improve the accuracy. However, I used Colab to train the models and PC for further works. Moreover, I have implemented this project using Keras and Pytorch. An overview of this project is available on YouTube.

Keras

Prerequisites

In Colab	In PC
• Python 3.6.9	• Python 3.7.9
• Tensorflow 2.4.0	• Tensorflow 2.4.0
• Keras 2.4.3	• Keras 2.4.3
• PIL 7.0.0	• PIL 8.0.0
• Numpy 1.19.4	• Numpy 1.19.4

Train A Particular Model

Part_1: In Colab

1. Upload the ModelName.ipynb file in colab.
2. Upload the Dataset.zip file in ModelName.ipynb notebook's Files section.
3. Run the notebook.
4. Two files can be found in the files section when training is complete. Download both ModelName.h5 and ModelName.json files and keep them in ./Saved Models directory.

Part_2: In PC

5. Extract Dataset.zip file.
6. Run LoadModel.ipynb and follow the notebook for further work.

Load A Pre-Trained Model

1. Extract Dataset.zip file.
2. Open LoadModel.ipynb in your PC.
3. Load the saved model.
4. Compile the loaded model.
5. Follow the notebook for further work.

Pytorch

License

Distributed under the MIT License. See LICENSE for more information.

Contact

Mehedi Hasan Bijoy - @Facebook - @LinkedIn