US 11,055,557 B2
Automated extraction of product attributes from images
Anirban Chatterjee, District-Howrah (IN); Bodhisattwa Prasad Majumder, Kolkata (IN); Gayatri Pal, Kodigehalli (IN); Rajesh Shreedhar Bhat, Kumta (IN); Sumanth S. Prabhu, Bangalore (IN); and Vignesh Selvaraj, Coimbatore (IN)
Assigned to Walmart Apollo, LLC, Bentonville, AR (US)
Filed by Walmart Apollo, LLC, Bentonville, AR (US)
Filed on May 17, 2018, as Appl. No. 15/982,675.
Claims priority of application No. 201841012953 (IN), filed on Apr. 5, 2018.
Prior Publication US 2019/0311210 A1, Oct. 10, 2019
Int. Cl. G06K 9/00 (2006.01); G06K 9/32 (2006.01); G06N 5/04 (2006.01); G06N 20/00 (2019.01); G06F 16/58 (2019.01); G06F 16/33 (2019.01); G06F 40/279 (2020.01)
CPC G06K 9/325 (2013.01) [G06F 16/3334 (2019.01); G06F 16/58 (2019.01); G06F 40/279 (2020.01); G06K 9/3241 (2013.01); G06N 5/046 (2013.01); G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A method for enhancing product descriptions, the method comprising:
obtaining at least a first product image associated with a product, the first product image including textual information about the product, the textual information being represented by at least one relevant portion of the first product image;
executing a text recognition module to recognize text in the first product image, wherein executing the text recognition module includes executing an attention layer that emphasizes an area of the first product image to receive greater visual attention and emphasize word character sequence in the area of the first product image;
processing the recognized text in the first product image using an end-to-end automated machine learning system (an ML system) for attribute tagging to determine one or more attributes associated with the product, wherein the attribute tagging includes assigning a token to each word of recognized text from the textual information, the token being either a starting token of an individual attribute, a continuing token of the individual attribute, or a disregard token if the word is unrelated to the individual attribute, and wherein the individual attribute is at least one of a term sequence or a character sequence; and
storing the one or more attributes associated with the product in a database used to generate an online description of the product.