US 12,260,623 B2
Training method and apparatus for image region segmentation model, and image region segmentation method and apparatus
Jun Zhang, Shenzhen (CN); Kuan Tian, Shenzhen (CN); Kezhou Yan, Shenzhen (CN); Jianhua Yao, Shenzhen (CN); and Xiao Han, Shenzhen (CN)
Assigned to TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Shenzhen (CN)
Filed by TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Shenzhen (CN)
Filed on Mar. 29, 2022, as Appl. No. 17/707,045.
Application 17/707,045 is a continuation of application No. PCT/CN2021/087128, filed on Apr. 14, 2021.
Claims priority of application No. 202010419791.7 (CN), filed on May 18, 2020.
Prior Publication US 2022/0222932 A1, Jul. 14, 2022
Int. Cl. G06K 9/00 (2022.01); G06T 7/11 (2017.01); G06T 7/194 (2017.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06V 10/7747 (2022.01) [G06T 7/11 (2017.01); G06T 7/194 (2017.01); G06V 10/82 (2022.01)] 17 Claims
OG exemplary drawing
 
1. A method for training an image segmentation model, the method comprising:
acquiring, by a device comprising a memory storing instructions and a processor in communication with the memory, a sample image set, each image of the sample image set having first annotation information, the first annotation information comprising an area proportion of a target region in the sample image;
generating, by the device, graph structure data corresponding to a sample image in the sample image set, the graph structure data comprising multiple nodes, and each node comprising at least one pixel in the sample image;
determining, by the device, second annotation information of each node according to the graph structure data and the first annotation information corresponding to the sample image by using a graph convolutional network model, a granularity of the second annotation information being smaller than a granularity of the first annotation information, the graph convolutional network model being a part of an image segmentation model, by:
obtaining a prediction result of each node according to the graph structure data by using the graph convolutional network model,
determining, according to the area proportion and a total quantity of nodes in the graph structure data, a first node quantity corresponding to the target region and a second node quantity corresponding to a background region, the background region being a region in the sample image except the target region, and
determining the second annotation information of each node according to the first node quantity, the second node quantity, and the prediction result; and
training, by the device, the image segmentation model according to the second annotation information.