US 12,293,480 B2
Method for reconstructing three-dimensional model, method for training three-dimensional reconstruction model, and apparatus
Yandan Zhao, Shenzhen (CN); Shuheng Lin, Shenzhen (CN); Xuan Cao, Shenzhen (CN); Yanhao Ge, Shenzhen (CN); Chengjie Wang, Shenzhen (CN); and Weijan Cao, Shenzhen (CN)
Assigned to Tencent Technology (Shenzhen) Company Limited, Shenzhen (CN)
Filed by Tencent Technology (Shenzhen) Company Limited, Shenzhen (CN)
Filed on Oct. 28, 2022, as Appl. No. 17/976,259.
Application 17/976,259 is a continuation of application No. PCT/CN2021/112089, filed on Aug. 11, 2021.
Claims priority of application No. 202010969615.0 (CN), filed on Sep. 15, 2020.
Prior Publication US 2023/0048906 A1, Feb. 16, 2023
Int. Cl. G06T 19/20 (2011.01); G06T 15/04 (2011.01); G06T 15/50 (2011.01)
CPC G06T 19/20 (2013.01) [G06T 15/04 (2013.01); G06T 15/503 (2013.01); G06T 2219/2021 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for reconstructing a three-dimensional model, performed by a computer device, the method comprising:
obtaining an image feature coefficient of an input image;
obtaining, based on the image feature coefficient, a texture of the input image, and a shape of the input image, a global feature map and an initial local feature map, the global feature map comprising a global texture feature map and a global shape feature map;
performing edge smoothing on the initial local feature map to obtain a target local feature map, the target local feature map comprising a local texture feature map and a local shape feature map;
modifying the target local feature map to be of a same size as the global feature map,
splicing the global feature map and the target local feature map based on the texture to obtain a target texture image, wherein the splicing comprises fusing pixels located at a same position in the global texture feature map and the local texture feature map;
splicing the global feature map and the target local feature map based on the shape to obtain a target shape image, wherein the splicing comprises fusing pixels located at a same position in the global shape feature map and the local shape map; and
performing three-dimensional model reconstruction based on the target texture image and the target shape image to obtain a target three-dimensional model.