US 12,130,887 B2
Semantic segmentation network structure generation method and apparatus, device, and storage medium
Peng Sun, Shenzhen (CN); and Jiaxiang Wu, Shenzhen (CN)
Assigned to TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Shenzhen (CN)
Filed by Tencent Technology (Shenzhen) Company Limited, Shenzhen (CN)
Filed on Oct. 29, 2021, as Appl. No. 17/515,180.
Application 17/515,180 is a continuation of application No. PCT/CN2020/114372, filed on Sep. 10, 2020.
Claims priority of application No. 201911102046.3 (CN), filed on Nov. 12, 2019.
Prior Publication US 2022/0051056 A1, Feb. 17, 2022
Int. Cl. G06K 9/62 (2022.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01); G06F 18/25 (2023.01); G06N 3/04 (2023.01); G06N 3/082 (2023.01); G06T 7/70 (2017.01); G06V 10/26 (2022.01); G06V 10/70 (2022.01); G06V 10/80 (2022.01); G06V 10/82 (2022.01); G06V 20/40 (2022.01); G06V 20/70 (2022.01)
CPC G06F 18/214 (2023.01) [G06F 18/217 (2023.01); G06F 18/253 (2023.01); G06N 3/04 (2013.01); G06N 3/082 (2013.01); G06T 7/70 (2017.01); G06V 10/267 (2022.01); G06V 10/70 (2022.01); G06V 10/806 (2022.01); G06V 10/82 (2022.01); G06V 20/41 (2022.01); G06V 20/70 (2022.01); G06T 2207/20084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A semantic segmentation network structure generation method performed by an electronic device, the semantic segmentation network structure comprising a super cell and an aggregation cell, and the method comprising:
generating a corresponding architectural parameter for cells that form the super cell in the semantic segmentation network structure;
optimizing the semantic segmentation network structure based on image samples, and removing a redundant cell from the super cell to which a target cell pertains, to obtain an improved semantic segmentation network structure, the target cell being a cell having a maximum architectural parameter among the cells;
performing, by the aggregation cell in the improved semantic segmentation network structure, feature fusion on an output of the super cell from which the redundant cell is removed, to obtain a fused feature map;
performing recognition processing on the fused feature map, to determine positions corresponding to objects that are in the image samples; and
training the improved semantic segmentation network structure based on the positions corresponding to the objects that are in the image samples and annotations corresponding to the image samples, to obtain a trained semantic segmentation network structure.