BaybayInsight is a deep learning-powered application for recognizing Baybayin characters — an ancient Filipino script — using Convolutional Neural Networks (CNNs). This project aims to help preserve and promote Filipino heritage by providing an automated tool that can interpret hand-drawn Baybayin characters with high accuracy.
Baybayin (also known as Alibata) is a pre-colonial Philippine writing system. With the rise of modern digital tools, this project aims to automate the recognition of Baybayin characters through a trained CNN model. This system enables users to draw characters, which are then classified and interpreted as digital text.
- Frontend: Flutter
- Machine Learning: Python, TensorFlow, Keras, NumPy
- Training Platform: Google Colab (with GPU)
- Dataset: Kaggle (by James Arnold Nogra)
- Total images: 10,000+
- Number of classes: 63 Baybayin characters
- Preprocessing:
- Resized and normalized image data
- Split into training, validation, and testing sets
The CNN model consists of:
- 3 Convolutional Blocks:
- Conv2D → BatchNormalization → ReLU → MaxPooling2D → Dropout
- GlobalAveragePooling2D
- Fully Connected Dense Layers
- Softmax Output Layer (63 classes)
The full model training workflow — including data preprocessing, CNN architecture, model training, evaluation, and visualization — is documented in this Google Colab notebook:
👉 Baybayin Character Recognition – Colab Notebook
You can open and run the notebook directly in your browser using Google Colab. GPU acceleration is recommended for faster training.
- Filter sizes: 3x3, 5x5, 7x7
- Channel sizes: 8–128
- Adam Optimizer + Cross-Entropy Loss
- Dropout Regularization
- Trained for 25 epochs
- Training Accuracy: 98.89%
- Validation Accuracy: 98.48%
- Visual metrics include:
- Epoch Accuracy Graph
- Confusion Matrix
- Clone the repository:
git clone https://github.com/drewsix9/BaybayInsight.git
cd BaybayInsight- Install Flutter dependencies:
flutter pub get- Launch the FLutter app:
flutter run👥 Contributors
- Dan Mathew Gamale
- Jennifer Garin
- Jessa Marie Ilogon
🌏 Acknowledgements
- Baybayin dataset by James Arnold Nogra on Kaggle
- Inspired by the cultural heritage of the Philippines