BrahmiGAN

This dataset was developed as part of our research project titled "A Novel GAN-Transformer Framework for Early Brahmi Script Generation and Recognition", published in Engineering and Applied Science Research (EASR), Vol. 53, No. 2, 2026. The dataset contains 21,195 synthetic images of 23 Early Brahmi letter classes, generated using BrahmiGAN — a novel GAN architecture incorporating residual connections, dual attention mechanisms, and multi-scale feature extraction.  Paper: doi.org If you use this dataset in your research, please cite our paper as follows: Surasinghe, P., & Thanikasalam, K. (2026). A novel GAN-transformer framework for early Brahmi script generation and recognition. Engineering and Applied Science Research, 53(2), 112–126. doi.org This dataset is provided solely for research and academic purposes, without any commercial intent. We encourage its use in future studies related to ancient script recognition, epigraphical analysis, data augmentation for low-resource scripts, and cultural heritage preservation.