US 12,112,500 B2
Apparatus for performing cylindrical convolutional neural network operation and apparatus and method for object recognition and viewpoint estimation using the same
Kwang Hoon Sohn, Seoul (KR); and Sung Hun Joung, Seoul (KR)
Assigned to INDUSTRY-ACADEMIC COOPERATION FOUNDATION, YONSEI UNIVERSITY, Seoul (KR)
Filed by INDUSTRY-ACADEMIC COOPERATION FOUNDATION, YONSEI UNIVERSITY, Seoul (KR)
Filed on Sep. 8, 2021, as Appl. No. 17/468,948.
Application 17/468,948 is a continuation of application No. PCT/KR2020/095103, filed on Aug. 14, 2020.
Claims priority of application No. 10-2020-0101612 (KR), filed on Aug. 13, 2020.
Prior Publication US 2022/0051433 A1, Feb. 17, 2022
Int. Cl. G06T 7/73 (2017.01); G06F 18/28 (2023.01); G06V 10/40 (2022.01)
CPC G06T 7/73 (2017.01) [G06F 18/28 (2023.01); G06V 10/40 (2022.01); G06T 2207/20084 (2013.01)] 18 Claims
OG exemplary drawing
 
1. An apparatus for cylindrical convolutional neural network operation, the apparatus comprising a cylindrical kernel having a structure in which a plurality of weight columns corresponding to a pre-designated angular unit interval are periodically repeated in units of 360 degrees, and
when an input image of a pre-designated size is applied, the apparatus performing a convolution operation on the number of weight columns corresponding to an angular range of a pre-designated size among the plurality of weight columns of the cylindrical kernel and the input image, to extract a plurality of view-specific feature vectors according to each angular unit interval,
wherein the cylindrical kernel is stored as a weight array of a two-dimensional structure in which a weight column corresponding to a pre-designated angle is disconnected and unfolded among the plurality of weight columns having a structure that is repeated at a cycle of 360 degrees according to a corresponding angle,
wherein the cylindrical kernel is stored as a plurality of weight arrays in which a pre-designated number of weight columns are combined adjacent to each other based on the weight column corresponding to each angle.