Authors: nian xie、liangwei wang、rongfu zheng、wentao li、qiang fu、jun he、zhiguang liu
Affiliation: Huawei, Noah's Ark LAB
Description: We developed a brand new synthetic OCR sample generation framework, which supports rich and diverse text styling. It provides a unified interface for basic text manipulation, like font, size, color, char space, etc. Even more, it offers special effects, such as 3D text, glowing text, curve text, engraved text, etc.
We trained some state of the art models on samples produced by our framework and achieve competitive performance.
Moreover, we made some modifications on SOTA models to further enhance our models’ performance.
The paper is in preparation