DigitDecoder

DigitDecoder is a PyTorch-powered handwritten digit recognizer. Draw or upload images, and the CNN model predicts digits (0–9) with high accuracy via a simple Streamlit interface. Perfect for learning, AI projects, or digitization tasks.

Repo URL: https://github.com/dpm24800/DigitDecoder
Notebook on Colab: Open in Colab
Deployed App: DigitDecoder

---

Features

• Recognizes handwritten digits (0–9) with high accuracy.
• Draw digits on a web-based canvas or upload images for prediction.
• Pre-trained CNN model for fast inference.
• Modular design: single or combined prediction features.

---

Files and Usage

Model Training

• Run the notebook digit-decoder.ipynb or the script digit-decoder.py to train the model.
• The model is trained on handwritten digit datasets (MNIST).

Prediction / Inference

• Both features (drawing + image upload): app-both.py
• Single feature (drawing only): app-drawer.py
• Single feature (image upload only): app-uploader.py

Run any of the apps with:

streamlit run <filename>.py

---

Screenshots & Demo

Drawing Digits:

Upload Image for Prediction:

Tip: You can try the live app at Streamlit URL to interact with the model in real-time.

---

Installation

1. Clone the repository:

git clone https://github.com/dpm24800/DigitDecoder.git
cd DigitDecoder

2. Install dependencies:

pip install -r requirements.txt

---

Technologies Used

• Python
• PyTorch & Torchvision
• Streamlit
• NumPy & PIL

---

License

Distributed under the MIT License. See LICENSE for details.

Contact

Dipak Pulami Magar – @dpm24800
Project Link: https://github.com/dpm24800/DigitDecoder