Abstract: The disclosure herein describes providing signature data of an input document. Text data of the input document is obtained (e.g., OCR data generated from image data) and a first set of signature fields are identified using signature key-value pairs of the text data. A first subset of signed signature fields and a first subset of unsigned signature fields are determined based on mapping to a set of predicted values. A second set of signature fields are determined using a region prediction model applied to image data of the input document. Region images associated with the first subset of unsigned signature fields and with second set of signature fields are obtained and a second set of signed signature fields and a second set of unsigned signature fields are determined using a signature recognition model. Signature output data is provided including signed signature fields and/or unsigned signature fields.

Abstract: Embodiments relate to a two-stage end-to-end text recognition system. The text recognition system includes a text detection stage and a text recognition stage. Images inputted to the text recognition system are provided to both the text detection stage and to the text recognition stage. The text detection stage detects text regions in the images and provides the detected regions to the text recognition stage. The text recognition stage is trained to perform geometric rectification on the text regions using the images. There is end-to-end alignment between the text detection stage and the text recognition stage. Additionally, the text detection stage and text recognition stage are each trained independent of the other.

Abstract: Embodiments relate to a two-stage end-to-end text recognition system. The text recognition system includes a text detection stage and a text recognition stage. Images inputted to the text recognition system are provided to both the text detection stage and to the text recognition stage. The text detection stage detects text regions in the images and provides the detected regions to the text recognition stage. The text recognition stage is trained to perform geometric rectification on the text regions using the images. There is end-to-end alignment between the text detection stage and the text recognition stage. Additionally, the text detection stage and text recognition stage are each trained independent of the other.