Digit Classifier

This is a handwritten digit classifier using a scaled-down VGG (TinyVGG) model. The model is trained on images from the MNIST dataset. The web framework used for this is Panel.

A version of this web application is deployed on Hugging Face Spaces as a quick demo. Please note that its performance may vary due to limited resource allocations (uses CPU basic). For a consistently smooth experience, it is recommended to run the application locally.

Recommended Installation Instructions

1) Create a New Python Environment

This environment should use Python >= 3.10.

2) Clone the digit-classifier-app Repo

Navigate to your desired parent directory and clone the repository:

git clone https://github.com/Jechen00/digit-classifier-app.git

3) Install Packages

In the digit-classifier-app repository run one of the following:

pip install -r requirements.txt

Alternatively, the packages can be installed manually like so:

pip install numpy==2.2.4
pip install matplotlib==3.10.1
pip install panel==1.4.5
pip install param==2.1.1
pip install plotly==6.0.1
pip install torch==2.6.0
pip install torchvision==0.21.0

Running the Application

In the digit-classifier-app repository, the web application can be ran locally with

panel serve app.py --port=66

The web application will be lauched at http://localhost:66/app.

VGG Reference Links:

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Networks Using Blocks (VGG)
• CNN Explainer