Description
Central to this dataset is the creation of “Lara”, a unique dataset comprised of 10,000 instances from “Bustan” by Saadi, aiming to encapsulate the calligraphic diversity of Nasta’liq. This dataset partitioned into an 80:20 training-to-testing ratio, enabled a rigorous validation of the framework, which demonstrated an unprecedented 98.8% accuracy in script recognition, significantly surpassing existing benchmarks.