Armenian-Handwriting-GAN

Summary

This repository implements an end-to-end system for generating synthetic Armenian handwritten text using an Auxiliary Classifier Wasserstein GAN with Gradient Penalty (AC-WGAN-GP). It addresses the scarcity of labeled Armenian handwriting datasets for OCR development by providing a data augmentation pipeline. The project includes a fair comparison with cVAE and DDPM models, a digraph-aware tokenizer for Armenian ligatures, and post-processing for realistic text blending on backgrounds.