US 12,380,689 B2
Managing occlusion in Siamese tracking using structured dropouts
Deepak Kumar Gupta, Amsterdam (NL); Efstratios Gavves, Amsterdam (NL); and Arnold Wilhelmus Maria Smeulders, Amsterdam (NL)
Assigned to QUALCOMM Incorporated, San Diego, CA (US)
Appl. No. 17/794,555
Filed by QUALCOMM Technologies, Inc., San Diego, CA (US)
PCT Filed Mar. 18, 2021, PCT No. PCT/US2021/023045
§ 371(c)(1), (2) Date Jul. 21, 2022,
PCT Pub. No. WO2021/188843, PCT Pub. Date Sep. 23, 2021.
Claims priority of application No. 20200100142 (GR), filed on Mar. 18, 2020.
Prior Publication US 2023/0070439 A1, Mar. 9, 2023
Int. Cl. G06V 10/82 (2022.01); G06T 5/20 (2006.01); G06T 7/73 (2017.01)
CPC G06V 10/82 (2022.01) [G06T 5/20 (2013.01); G06T 7/73 (2017.01); G06T 2207/20076 (2013.01); G06T 2207/20084 (2013.01)] 28 Claims
OG exemplary drawing
 
1. A method, comprising:
receiving a target image of an object of interest;
extracting latent space features of the target image;
modifying the latent space features of the target image at a forward pass of a neural network using a structured dropout to model one or more appearance variations of the target image;
predicting, at the forward pass, a location of the object of interest in a search image based on the modified latent space features; and
identifying the location of the object of interest by aggregating predicted locations from the forward pass.