US 12,086,728 B2
Form structure extraction by predicting associations
Milan Aggarwal, San Jose, CA (US); Mausoom Sarkar, San Jose, CA (US); and Balaji Krishnamurthy, San Jose, CA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Apr. 18, 2023, as Appl. No. 18/135,948.
Application 18/135,948 is a division of application No. 16/904,263, filed on Jun. 17, 2020, granted, now 11,657,306.
Prior Publication US 2023/0267345 A1, Aug. 24, 2023
Int. Cl. G06N 5/04 (2023.01); G06N 3/08 (2023.01); G06N 20/00 (2019.01); G06N 20/10 (2019.01); G06V 10/82 (2022.01)
CPC G06N 5/04 (2013.01) [G06N 3/08 (2013.01); G06N 20/00 (2019.01); G06N 20/10 (2019.01); G06V 10/82 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
accessing textruns and widgets extracted from a form;
training, using binary cross entropy loss, a first set of prediction models to predict textblocks based on the textruns and the widgets, each textblock of the textblocks comprising a respective group of textruns;
applying the first set of prediction models to the textruns to determine the textblocks from the textruns and the widgets;
training, using cross entropy loss and binary cross entropy loss, a second set of prediction models to predict form groups based on the textblocks and the widgets, each form group of the form groups comprising a combination of textblocks or widgets;
applying the second set of prediction models to the textblocks and the widgets to determine the form groups; and
generating a reflowable form based on the form and comprising the form groups.