Title:
POSE RECOGNITION MODEL TRAINING METHOD AND APPARATUS, POSE RECOGNITION METHOD, AND TERMINAL DEVICE
Document Type and Number:
WIPO Patent Application WO/2022/127494
Kind Code:
A1
Abstract:
Disclosed in embodiments of the present application are a pose recognition model training method and apparatus, a pose recognition method, and a terminal device. The training method comprises: acquiring an original RGB image; extracting, from the original RGB image, by using a preset pose point-line graph extraction method, a pose point-line graph composed of coordinates of a preset number of pose key points; drawing the pose point-line graph on a background board in a target color different from a predetermined background board color, so as to determine a training RGB image; and training a pose recognition model by using a plurality of training RGB images until a loss function corresponding to the pose recognition model converges. According to the technical solution of the present application, a pose recognition model can be trained by using a small number of training RGB images, and the time of training the pose recognition model is reduced.
More Like This:
Inventors:
LIN CANRAN (CN)
GUO MIAOCHEN (CN)
CHENG JUN (CN)
PANG JIANXIN (CN)
GUO MIAOCHEN (CN)
CHENG JUN (CN)
PANG JIANXIN (CN)
Application Number:
PCT/CN2021/131148
Publication Date:
June 23, 2022
Filing Date:
November 17, 2021
Export Citation:
Assignee:
UBTECH ROBOTICS CORP LTD (CN)
International Classes:
G06K9/62; G06T7/73; G06N3/04; G06N3/08; G06T15/00
Foreign References:
| CN112489129A | 2021-03-12 | |||
| CN111274998A | 2020-06-12 | |||
| CN107886089A | 2018-04-06 | |||
| CN108846365A | 2018-11-20 | |||
| US20200160046A1 | 2020-05-21 |
Attorney, Agent or Firm:
CHOFN INTELLECTUAL PROPERTY (CN)
Download PDF: