Fetching the paper…

TextOCR: Towards large-scale end-to-end reasoning for arbitrary-shaped scene text · Around