Official data of the paper "What’s Wrong with Synthetic Data for Scene Text Recognition? A Strong Synthetic Engine with Diverse Simulations and Self-Evolution".
Scene Text Recognition (STR) relies critically on large-scale, high-quality training data. While synthetic data provides a cost-effective alternative to manually annotated real data, existing rendering-based synthetic datasets suffer… See the full description on the dataset page:
https://huggingface.co/datasets/Yesianrohn/UnionST.
