US 11,815,942 B2
Systems and methods for executing attention-based object searches on images using neural network architectures
Theban Stanley, San Bruno, CA (US); Nihar Vanjara, San Jose, CA (US); Yanxin Pan, Sunnyvale, CA (US); Abon Chaudhuri, Sunnyvale, CA (US); and Nikash Walia, San Jose, CA (US)
Assigned to WALMART APOLLO, LLC, Bentonville, AR (US)
Filed by Walmart Apollo, LLC, Bentonville, AR (US)
Filed on Mar. 14, 2022, as Appl. No. 17/694,517.
Claims priority of provisional application 63/160,856, filed on Mar. 14, 2021.
Prior Publication US 2022/0292129 A1, Sep. 15, 2022
Int. Cl. G06F 16/53 (2019.01); G06N 3/08 (2023.01)
CPC G06F 16/53 (2019.01) [G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable storage devices storing computing instructions configured to run on the one or more processors and perform acts of:
receiving, at a neural network architecture, a query image comprising at least one target object;
receiving, at the neural network architecture, at least one candidate image;
generating, using a region proposal network of the neural network architecture, a plurality of proposals based on the at least one candidate image;
selecting, using a proposal selection model of the neural network architecture, a portion of the plurality of proposals to produce a reduced proposal set;
generating a query embedding corresponding to the query image;
generating candidate embeddings corresponding to the portion of the plurality of proposals included in the reduced proposal set;
computing similarity scores for the portion of the plurality of proposals included in the reduced proposal set based on comparisons of the query embedding to each of the candidate embeddings; and
comparing the similarity scores to a threshold to determine if the at least one candidate image comprises the at least one target object.