Mashtots-Dataset
TyomPapiyan/Mashtots-Dataset
CNN for handwritten Armenian character recognition (78 classes) — Kaggle Mashtots Dataset competition, ~0.94 validation accuracy
handwriting-recognitionconvolutional-neural-networkkerascomputer-visionimage-classificationkaggle-competitionjupyter-notebook
Summary
A complete Jupyter Notebook project for handwritten Armenian character recognition using a CNN built from scratch in Keras. It trains on the Kaggle Mashtots Dataset (78 classes, ~70k grayscale glyphs), implements data loading, preprocessing, model architecture, training with callbacks, test-time augmentation, and generates competition submissions. The project is well-documented and serves as a practical, end-to-end deep learning pipeline.