SELECT: Enhancing Scene Text Recognition with Error Detection

This research focuses on improving the accuracy of scene text recognition by identifying and mitigating label errors in real-world datasets. The paper's contribution is in developing a method (SELECT) to address a crucial problem in training robust text recognition models.

• •Addresses the problem of noisy labels in scene text datasets.
• •Proposes a method named SELECT for error detection.
• •Contributes to improved accuracy in text recognition models.