Digit Recognition Project

This is a Python 3.11 project which implements digit recognition using a Convolutional Neural Network (CNN). The dataset utilized is MNIST. It contains handwritten digit images for training and validating the model's digit classification capabilities.

Project Structure

├── data/MNIST/raw
│   └── handwritten images used for training and testing.
│
├── frontend/digit_recog_frontend
│   └── frontend implementation of the model. It is a react projecct. 
│       See the README file in this folder for details. 
│                            
├── export.py (Converts a trained PyTorch model into an ONNX model.)
│
├── Main.ipynb (The Jupyter Notebook code. The model was firstly written in Jupyter, 
│               then converted into python code. It is equivalent to main.py.)
│                             
├── main.py (Contains the implementation of the CNN, the training, and the testing. )
│
├── model.py (The ONNX model generated by export.py.)
│
└── model_weights.pth (Trained PyTorch model params will be saved as model_weights.pth.)

Dependencies

• torch
• torchvision
• numpy

You can install the required dependencies using the following command

pip install torch torchvision numpy

Running the Project

1. Install Dependencies: Ensure that you have installed the required dependencies.
2. Run the Code<\b>: Execute the main.py file in the project's root directory.
3. Save Model<\b>: The result parameters of the model in the training will be saved in model_weights.pth. Run the export.py file to export the model as onnx.

Project Overview

Model Structure

The project utilizes a simple Convolutional Neural Network named CNN in main.py.

class CNN(nn.module):
    # ...(Model architecture)

Training and Testing

The training and testing functions are defined in main.py after the definition of the CNN model

def train(epoch):
    #... (Training function)
    
def test():
    #... (Testing function)

Postscript

The accuracy rate of running the test image dataset is around 98%.

However, it demonstrates difficulty in distinguishing 7 and 1 in real testing.