US 12,169,882 B2
Learning dense correspondences for images
Sifei Liu, Santa Clara, CA (US); Jiteng Mu, La Jolla, CA (US); Shalini De Mello, San Francisco, CA (US); Zhiding Yu, Santa Clara, CA (US); and Jan Kautz, Lexington, MA (US)
Assigned to NVIDIA Corporation, Santa Clara, CA (US)
Filed by NVIDIA Corporation, Santa Clara, CA (US)
Filed on Sep. 1, 2022, as Appl. No. 17/929,182.
Claims priority of provisional application 63/306,773, filed on Feb. 4, 2022.
Prior Publication US 2023/0252692 A1, Aug. 10, 2023
Int. Cl. G06T 11/00 (2006.01); G06T 3/18 (2024.01)
CPC G06T 11/001 (2013.01) [G06T 3/18 (2024.01)] 20 Claims
OG exemplary drawing
 
1. A method of generating correspondences for images, comprising:
warping a shared coordinate map, according to transformation parameters, to align with an encoded structure to produce a correspondence map for an image; and
synthesizing, by a modulated generator comprising a sequence of processing layers and using generator parameters, the image according to the correspondence map and an encoded texture, wherein the correspondence map is directly combined with features output by one or more of the processing layers and an object in the image comprises appearance details according to the encoded texture and component shapes controlled by the encoded structure.