The advancement of text shape representations towards compactness has
enhanced text detection and spotting performance, but at a high annotation
cost. Current models use single-point annotations to reduce costs, yet they
lack sufficient localization information for downstream applicati