Data processing method and apparatus

Patent metadata

Field	Value
Publication number	US-2025157071-A1
Application number	US-202519024815-A
Country	US
Kind code	A1
Filing date	Jan 16, 2025
Priority date	Jul 20, 2022
Publication date	May 15, 2025
Grant date	—

Abstract

This disclosure provides data processing methods and devices relating to artificial intelligence. In an implementation, a method includes: processing a target image by using a first pose recognition model to obtain first pose information of a target object in the target image, processing the target image by using a second pose recognition model to obtain second pose information of the target object in the target image, and constructing a loss based on the first pose information, the second pose information, the two-dimensional projection information, and a corresponding annotation.

First claim

1 . A data processing method, comprising:
obtaining a target image; processing the target image by using a first pose recognition model, to obtain first pose information of a target object in the target image; processing the target image by using a second pose recognition model, to obtain second pose information of the target object in the target image, wherein the first pose information and the second pose information describe a three-dimensional pose of the target object, and wherein the second pose information determines two-dimensional projection information of a predicted pose of the target object; and constructing a loss for updating the second pose recognition model based on the first pose information, the second pose information, the two-dimensional projection information, and a corresponding annotation.

2 . The method according to claim 1 , wherein the first pose recognition model is obtained through training based on a loss constructed based on output pose information and a corresponding annotation.

3 . The method according to claim 1 , wherein
first body shape information of the target object in the target image is further obtained by processing the target image by using the first pose recognition model; second body shape information of the target object in the target image is further obtained by processing the target image by using the second pose recognition model; and wherein the constructing a loss further comprises: constructing the loss based on the first body shape information and the second body shape information.

4 . The method according to claim 1 , wherein the target image is an image area in which the target object is located in an original image, and the two-dimensional projection information is represented as a location of a two-dimensional projection of the predicted pose in the original image.

5 . The method according to claim 1 , wherein the target object is a character.

6 . The method according to claim 1 , wherein the method further comprises:
processing the target image by using an updated second pose recognition model, to obtain third pose information of the target object in the target image, wherein the third pose information determines a pose of the target object.

7 . The method according to claim 6 , wherein the method further comprises:
sending, to user equipment, the updated second pose recognition model or the pose of the target object obtained by processing the target image by using the updated second pose recognition model.

8 . The method according to claim 1 , wherein the annotation is a manual advance annotation, or is obtained by processing the target image by using a pre-trained model.

9 . A training device, comprising at least one processor and a memory coupled to the at least one processor, wherein the memory stores instructions for execution by the at least one processor to:
obtain a target image; process the target image by using a first pose recognition model, to obtain first pose information of a target object in the target image; process the target image by using a second pose recognition model, to obtain second pose information of the target object in the target image, wherein the first pose information and the second pose information describe a three-dimensional pose of the target object, and wherein the second pose information determines two-dimensional projection information of a predicted pose of the target object; and construct a loss for updating the second pose recognition model based on the first pose information, the second pose information, the two-dimensional projection information, and a corresponding annotation.

10 . The device according to claim 9 , wherein the first pose recognition model is obtained through training based on a loss constructed based on output pose information and a corresponding annotation.

11 . The device according to claim 9 , wherein
first body shape information of the target object in the target image is further obtained by processing the target image by using the first pose recognition model; second body shape information of the target object in the target image is further obtained by processing the target image by using the second pose recognition model; and wherein the constructing a loss further comprises: constructing the loss based on the first body shape information, and the second body shape information.

12 . The device according to claim 9 , wherein the target image is an image area in which the target object is located in an original image, and the two-dimensional projection information is represented as a location of a two-dimensional projection of the predicted pose in the original image.

13 . The device according to claim 9 , wherein the target object is a character.

14 . The device according to claim 9 , wherein the annotation is a manual advance annotation, or is obtained by processing the target image by using a pre-trained model.

15 . A computer program product, comprising computer-readable instructions, wherein the computer-readable instructions, when executed by a computer device, instruct the computer device to:
obtain a target image; process the target image by using a first pose recognition model, to obtain first pose information of a target object in the target image; process the target image by using a second pose recognition model, to obtain second pose information of the target object in the target image, wherein the first pose information and the second pose information describe a three-dimensional pose of the target object, and wherein the second pose information determines two-dimensional projection information of a predicted pose of the target object; and construct a loss for updating the second pose recognition model based on the first pose information, the second pose information, the two-dimensional projection information, and a corresponding annotation.

16 . The computer program product according to claim 15 , wherein the first pose recognition model is obtained through training based on a loss constructed based on output pose information and a corresponding annotation.

17 . The computer program product according to claim 15 , wherein
first body shape information of the target object in the target image is further obtained by processing the target image by using the first pose recognition model; second body shape information of the target object in the target image is further obtained by processing the target image by using the second pose recognition model; and wherein the constructing a loss further comprises: constructing the loss based on the first body shape information, and the second body shape information.

18 . The computer program product according to claim 15 , wherein the target image is an image area in which the target object is located in an original image, and the two-dimensional projection information is represented as a location of a two-dimensional projection of the predicted pose in the original image.

19 . The computer program product according to claim 15 , wherein the target object is a character.

20 . The computer program product according to claim 15 , wherein the annotation is a manual advance annotation, or is obtained by processing the target image by using a pre-trained model.

Assignees

• Huawei Tech Co Ltd

Inventors

• Li Zhihao
• Liu Jianzhuang
• Xu Songcen

Classifications

• G06T7/73

  using feature-based methods · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/30196

  Human being; Person · CPC title

• G06T7/60

  Analysis of geometric attributes · CPC title