SA-Text Image Text Dataset

SA-Text is a large-scale benchmark dataset of high-quality scene images released by the Korea Advanced Institute of Science and Technology and Korea University, designed for the task of text-aware image restoration (TAIR). The related paper results are:Text-Aware Image Restoration with Diffusion Models".

The dataset contains 105,330 high-resolution scene images with polygon-level text annotations, which can accurately describe the location and shape of text in the image, providing precise supervision information for the TAIR task, enabling the model to better understand the location and structure of text in the image.