公开(公告)号：US20210225069A1

公开(公告)日：2021-07-22

申请号：US17021218

申请日：2020-09-15

Applicant: BEIJING BAIDU NETCOM SCIENCE AND TECHNOLOGY CO., LTD.

Inventor： Qingyue MENG ,  Chen ZHAO

IPC: G06T17/00 ,  G06N20/00

Abstract: A method and apparatus for generating a 3D joint point regression model are provided. An embodiment of the method includes: acquiring a sample image with a 2D label and a sample image with a 3D label; training part of channels of an output layer of a basic 3D joint point regression model, with the sample image with the 2D label as a first input, and with a joint point heat map set corresponding to the 2D label as a first expected output; and training all of the channels of the output layer, with the sample image with the 3D label as a second input, with a joint point heat map set corresponding to the 3D label as a first part output of a second expected output and with a joint point depth information map set corresponding to the 3D label as a second part output of the second expected output.