US 11,790,550 B2
Learnable cost volume for determining pixel correspondence
Taihong Xiao, Merced, CA (US); Deqing Sun, Cambridge, MA (US); Ming-Hsuan Yang, Cupertino, CA (US); Qifei Wang, Sunnyvale, CA (US); and Jinwei Yuan, Sunnyvale, CA (US)
Assigned to Google LLC, Mountain View, CA (US)
Appl. No. 17/292,647
Filed by Google LLC
PCT Filed Jul. 8, 2020, PCT No. PCT/US2020/041258
§ 371(c)(1), (2) Date May 10, 2021,
PCT Pub. No. WO2022/010476, PCT Pub. Date Jan. 13, 2022.
Prior Publication US 2022/0189051 A1, Jun. 16, 2022
Int. Cl. G06K 9/00 (2022.01); G06T 7/593 (2017.01); G06T 7/215 (2017.01)
CPC G06T 7/593 (2017.01) [G06T 7/215 (2017.01); G06T 2207/10012 (2013.01); G06T 2207/20081 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
obtaining (i) a first plurality of feature vectors associated with a first image and (ii) a second plurality of feature vectors associated with a second image;
generating a plurality of transformed feature vectors by transforming each respective feature vector of the first plurality of feature vectors by a kernel matrix trained to define an elliptical inner product space;
generating a cost volume by determining, for each respective transformed feature vector of the plurality of transformed feature vectors, a plurality of inner products, wherein each respective inner product of the plurality of inner products is between the respective transformed feature vector and a corresponding candidate feature vector of a corresponding subset of the second plurality of feature vectors; and
determining, based on the cost volume, a pixel correspondence between the first image and the second image.