US 12,380,611 B2
Image generation using surface-based neural synthesis
Iason Kokkinos, London (GB); Georgios Papandreou, London (GB); and Riza Alp Guler, London (GB)
Assigned to Snap Inc., Santa Monica, CA (US)
Filed by Snap Inc., Santa Monica, CA (US)
Filed on Jul. 15, 2022, as Appl. No. 17/812,864.
Application 17/812,864 is a continuation of application No. 16/949,773, filed on Nov. 13, 2020, granted, now 11,430,247.
Claims priority of provisional application 62/936,328, filed on Nov. 15, 2019.
Prior Publication US 2022/0375247 A1, Nov. 24, 2022
Int. Cl. G06V 40/10 (2022.01); G06F 17/18 (2006.01); G06F 18/22 (2023.01); G06F 18/25 (2023.01); G06T 5/50 (2006.01); G06T 11/00 (2006.01); G06V 10/74 (2022.01); G06V 10/80 (2022.01); G06V 20/64 (2022.01)
CPC G06T 11/001 (2013.01) [G06F 17/18 (2013.01); G06F 18/22 (2023.01); G06F 18/253 (2023.01); G06T 5/50 (2013.01); G06V 10/74 (2022.01); G06V 10/806 (2022.01); G06V 20/647 (2022.01); G06V 40/103 (2022.01); G06T 2207/20084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
receiving a two-dimensional (2D) input image depicting a person in a first pose;
receiving a target image corresponding to a second pose;
processing the 2D input image to generate soft feature pooling features, the soft feature pooling features comprising a matrix of a collection of features extracted from the 2D input image that are arranged based on expected locations of different joints;
processing the target image to generate target soft intrinsic distances, the target soft intrinsic distances representing a likelihood that pixel values in the 2D input image correspond to a particular joint in the target image; and
decoding features of the 2D input image based on a combination of the soft feature pooling features and target soft intrinsic distances to generate a decoded image that depicts the person in the second pose.