US 12,367,705 B2
Occlusion detection
Dianbo Li, Shanghai (CN)
Assigned to Shanghai Bilibili Technology Co., Ltd., Shanghai (CN)
Filed by Shanghai Bilibili Technology Co., Ltd., Shanghai (CN)
Filed on Mar. 21, 2023, as Appl. No. 18/124,166.
Application 18/124,166 is a continuation of application No. PCT/CN2021/112308, filed on Aug. 12, 2021.
Claims priority of application No. 202011111254.2 (CN), filed on Oct. 16, 2020.
Prior Publication US 2023/0237841 A1, Jul. 27, 2023
Int. Cl. G06V 40/16 (2022.01); G06V 10/26 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06V 40/171 (2022.01) [G06V 10/26 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A method, comprising: constructing a plurality of pieces of training sample data, wherein the training sample data comprises a first face image added with an occlusion object, coordinate values of a first key point in the first face image, and occlusion information of the first key point; and using the first face image as input data, and using the coordinate values of the first key point and the occlusion information of the first key point as output data, to train an occlusion detection model, so that the occlusion detection model outputs, based on any input second face image, coordinate values of a second key point comprised in the second face image and an occlusion probability of the second key point; and wherein the constructing a plurality of pieces of training sample data comprises: obtaining an original face image that does not comprise the occlusion object; obtaining an original occlusion object image that comprises the occlusion object, and extracting a target occlusion object from the original occlusion object image; synthesizing the target occlusion object and the face image, to obtain the first face image added with the occlusion object; and recording the coordinate values of the first key point in the first face image and occlusion information of each first key point.