US 12,340,564 B2
Model generation method and apparatus, object detection method and apparatus, device, and storage medium
Yaozu An, Beijing (CN); Xinyu Xu, Beijing (CN); and Qi Kong, Beijing (CN)
Assigned to Beijing Jingdong Qianshi Technology Co., Ltd., Beijing (CN)
Appl. No. 17/912,342
Filed by Beijing Jingdong Qianshi Technology Co., Ltd., Beijing (CN)
PCT Filed Mar. 9, 2021, PCT No. PCT/CN2021/079690
§ 371(c)(1), (2) Date Sep. 16, 2022,
PCT Pub. No. WO2021/185121, PCT Pub. Date Sep. 23, 2021.
Claims priority of application No. 202010188303.6 (CN), filed on Mar. 17, 2020.
Prior Publication US 2023/0131518 A1, Apr. 27, 2023
Int. Cl. G06V 10/774 (2022.01); G06N 3/082 (2023.01); G06V 10/82 (2022.01); G06V 20/58 (2022.01); G06V 20/56 (2022.01)
CPC G06V 10/774 (2022.01) [G06N 3/082 (2013.01); G06V 10/82 (2022.01); G06V 20/58 (2022.01); G06V 20/56 (2022.01)] 19 Claims
OG exemplary drawing
 
1. A model generation method, performed by a model generation apparatus, wherein the model generation apparatus comprises a first processor, and a first memory configured to store a first program; wherein the first program, when executed by the model generation apparatus, causes the first processor to perform the model generation method;
the model generation method comprising:
acquiring, by the first processor, a plurality of scaling coefficients of a batch normalization layer in an initially-trained intermediate detection model, wherein the intermediate detection model is obtained by training an original detection model based on a plurality of training samples, and each of the plurality of training samples comprises a sample image and a sample annotation result of a known object in the sample image;
screening, by the first processor, a first coefficient from the plurality of scaling coefficients according to values of the plurality of scaling coefficients; and
screening, by the first processor, a first channel corresponding to the first coefficient from a plurality of channels of the intermediate detection model, and performing, by the first processor, channel pruning on the first channel to generate an object detection model.