US 12,136,026 B2
Compact neural networks using condensed filters
Yingzhen Yang, Los Angeles, CA (US); Jianchao Yang, Los Angeles, CA (US); and Ning Xu, Irvine, CA (US)
Assigned to SNAP INC., Santa Monica, CA (US)
Filed by Snap Inc., Santa Monica, CA (US)
Filed on Jul. 17, 2023, as Appl. No. 18/222,649.
Application 18/222,649 is a continuation of application No. 16/949,994, filed on Nov. 23, 2020, granted, now 11,763,130.
Application 16/949,994 is a continuation of application No. 16/155,656, filed on Oct. 9, 2018, granted, now 10,872,292.
Claims priority of provisional application 62/569,907, filed on Oct. 9, 2017.
Prior Publication US 2023/0359859 A1, Nov. 9, 2023
Int. Cl. G06K 9/00 (2022.01); G06N 3/04 (2023.01); G06T 5/20 (2006.01); G06T 7/10 (2017.01)
CPC G06N 3/04 (2013.01) [G06T 5/20 (2013.01); G06T 7/10 (2017.01)] 17 Claims
OG exemplary drawing
 
1. A mobile device comprising:
one or more processors of a machine; and
one or more memories storing instructions that, when executed by the one or more processors, cause the machine to perform operations comprising:
generating, from a compound neural network filter comprising a plurality of weights, a plurality of filters for a convolution layer in a convolutional neural network, wherein the plurality of filters share weights of the plurality of weights, wherein the convolutional neural network comprises a plurality of convolution layers, each convolution layer having a corresponding compound neural network filter comprising a corresponding plurality of weights configured to enable generating a corresponding plurality of filters;
applying the plurality of filters to an image using the convolutional neural network to generate a modified image; and
causing the modified image to be stored.