KeneanDita/Fidel-Vision

A Deep Learning + Streamlit web app for recognizing handwritten Amharic fidel characters (34 root groups × 7 orders). It uses a CNN model built with TensorFlow/Keras, trained on a custom dataset of handwritten Amharic letters, and serves predictions through a Streamlit interface. # Amharic Handwritten Character Recognition A **Deep Learning + Streamlit web app** for recognizing handwritten **Amharic fidel characters** (34 root groups × 7 orders). It uses a **CNN model built with TensorFlow/Keras**, trained on a custom dataset of handwritten Amharic letters, and serves predictions through a Streamlit interface. Author : Kenean Dita ### Acknowledgements * Dataset: Custom handwritten Amharic letters. * Frameworks: TensorFlow, Streamlit. ## Sample outputs - Using the canvas inside the webapp - By uploading an image to the webapp ## Features * **CNN model** trained on 32×32 grayscale images of handwritten fidel. * Supports **34 × 7 = 238 Amharic characters**. * **Streamlit web app** for interactive testing. * Dataset images stored in a single `data/` folder. * **Label mapping** from dataset class names → real Amharic letters. ## Project Structure ```PS . ├── data/ # Dataset (images of letters) ├── Models/ │ ├── amharic_cnn.h5 # Trained CNN model │ ├── class_names.npy # Encoded class labels ├── UI.py # Streamlit web app ├── training_notebook.ipynb # Training notebook ├── requirements.txt # Python dependencies ├── License ├── Dockerfile # For containerizaiton purposes ├── README.md # Project documentation ``` ## Usage ### Environment Setup Clone the repo: ```bash git clone github.com cd .\fidel-vision ``` Create a virtual envionment and install dependencies ```bash python -m venv env .\env\source\activate # for windows source env\source\activate # for linux/mac pip install -r requirements.txt ``` ### Train the model (optional) If you want to retrain: ```bash jupyter notebook train.ipynb ``` ### Run the Streamlit app ```bash streamlit run .\UI.py ``` * Upload a handwritten Amharic character image or draw using the second tab as an option. * The app preprocesses it → feeds it into the CNN → predicts th …