Handwritten Digit Recognition App

Description:

This project implements a web application using Flask and PyTorch to predict the digit drawn by users on a canvas element. It provides users with a visually intuitive way to interact with a machine learning model for digit recognition.

Key Features:

• User-friendly canvas for drawing: Users can easily draw digits on a designated canvas area within the web interface.
• Real-time prediction: As users draw, the app dynamically sends the canvas data to the Flask server for prediction.
• Probability distribution: The predicted digit is displayed along with its corresponding probability distribution, offering insights into the model's confidence level.
• Clear instructions: The application provides clear instructions to guide users on how to interact with the canvas and interpret the results.

Technologies Used:

• Flask: A lightweight web framework for building web applications in Python.
• PyTorch: A popular deep learning framework with GPU acceleration capabilities.
• HTML/CSS/JavaScript: Used for creating the interactive canvas interface and displaying predictions.

Requirements:

Python 3.8 or higher PyTorch 1.7.1 or higher

Installation:

1. Clone the repository:

git clone https://github.com/longvh-dev/handwritten_recognition_app.git

2. Install dependencies (Using conda enviroment)

conda env create -f environment.yml -n <env_name>
conda activate <env_name>

Remove the <env_name> and replace it with your desired environment name.

3. Usage:

Start the Flask application from the root directory of the project by running the following command in the terminal

python server/app.py

Open your web browser and navigate to http://127.0.0.1:5000/ (or the specified port in app.py).

Draw a digit on the canvas: Use your mouse or touch screen to create a digit within the designated area.

Wait for the prediction: Once you stop drawing, the application will send the canvas data to the server. This triggers the prediction and displays the predicted digit and its probability distribution on the page.

Train Model:

The project use a CNN model with 2 layer for digit recognition. If you want to using different model, you can change the model in the train/models/cnn.py file and update the app.py file to load the new model.

First, you need to go to the train folder and run the train.sh file to train the model.

bash train.sh

If you want to change parameters (learning_rate, num_epochs, device) for training, you can change the train.sh file.

Contribution:

Contributions are welcome! Feel free to fork the repository, make changes, and create pull requests to improve the project.

License:

This project is licensed under the MIT License. For more information, please refer to the LICENSE file.