US 12,437,502 B2
Method for generating objective function, apparatus, electronic device and computer readable medium
Yi He, Beijing (CN); and Changhu Wang, Beijing (CN)
Appl. No. 18/005,081
Filed by Beijing Bytedance Network Technology Co., Ltd., Beijing (CN)
PCT Filed May 26, 2021, PCT No. PCT/CN2021/096144
§ 371(c)(1), (2) Date Jan. 11, 2023,
PCT Pub. No. WO2022/012178, PCT Pub. Date Jan. 20, 2022.
Claims priority of application No. 202010685574.2 (CN), filed on Jul. 16, 2020.
Prior Publication US 2023/0281956 A1, Sep. 7, 2023
Int. Cl. G06V 10/72 (2022.01); G06F 18/15 (2023.01); G06F 18/21 (2023.01); G06F 18/213 (2023.01); G06F 18/22 (2023.01); G06N 20/00 (2019.01); G06T 5/60 (2024.01); G06T 7/10 (2017.01); G06T 7/13 (2017.01); G06T 9/00 (2006.01); G06V 10/32 (2022.01); G06V 10/40 (2022.01); G06V 10/44 (2022.01); G06V 10/74 (2022.01); G06V 10/75 (2022.01); G06V 10/77 (2022.01); G06V 30/18 (2022.01); G06V 30/182 (2022.01); G06V 40/16 (2022.01); H04L 9/06 (2006.01); G06F 16/13 (2019.01); G06F 16/22 (2019.01); G06F 16/31 (2019.01); G06F 16/901 (2019.01); H04L 7/00 (2006.01); H04N 21/234 (2011.01)
CPC G06V 10/72 (2022.01) [G06F 18/15 (2023.01); G06F 18/21 (2023.01); G06F 18/213 (2023.01); G06F 18/22 (2023.01); G06N 20/00 (2019.01); G06T 5/60 (2024.01); G06T 7/10 (2017.01); G06T 7/13 (2017.01); G06T 9/00 (2013.01); G06V 10/32 (2022.01); G06V 10/40 (2022.01); G06V 10/44 (2022.01); G06V 10/74 (2022.01); G06V 10/752 (2022.01); G06V 10/757 (2022.01); G06V 10/758 (2022.01); G06V 10/761 (2022.01); G06V 10/7715 (2022.01); G06V 30/18 (2022.01); G06V 30/1801 (2022.01); G06V 30/182 (2022.01); G06V 40/161 (2022.01); G06V 40/168 (2022.01); G06V 40/171 (2022.01); H04L 9/0643 (2013.01); G06F 16/137 (2019.01); G06F 16/2255 (2019.01); G06F 16/325 (2019.01); G06F 16/9014 (2019.01); G06F 2101/14 (2013.01); G06F 2218/08 (2023.01); G06F 2218/12 (2023.01); G06T 2207/20081 (2013.01); H04L 7/0087 (2013.01); H04N 21/23418 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A method for generating a target function comprising:
performing normalization processing on a vector corresponding to each pixel in a target contour feature map set to generate a target vector, so as to obtain a target vector set, wherein the target contour feature map set is obtained by extracting features of a target image set over a feature extraction network;
generating hash coding corresponding to each vector in the target vector set, to obtain a uniform hash coding set;
determining a prior probability of each hash coding in the uniform hash coding set;
generating a target function based on an entropy of the prior probability, the prior probability represents distribution of image features; and
optimizing the feature extraction network based on the target function to obtain a trained feature extraction network.