US 12,136,155 B2
System and method for photorealistic image synthesis using unsupervised semantic feature disentanglement
Yutong Zheng, Pittsburgh, PA (US); Marios Savvides, Pittsburgh, PA (US); and Yu Kai Huang, Pittsburgh, PA (US)
Assigned to Carnegie Mellon University, Pittsburgh, PA (US)
Appl. No. 18/269,721
Filed by CARNEGIE MELLON UNIVERSITY, Pittsburgh, PA (US)
PCT Filed Feb. 9, 2022, PCT No. PCT/US2022/015797
§ 371(c)(1), (2) Date Jun. 26, 2023,
PCT Pub. No. WO2022/173814, PCT Pub. Date Aug. 18, 2022.
Claims priority of provisional application 63/149,375, filed on Feb. 15, 2021.
Prior Publication US 2024/0062441 A1, Feb. 22, 2024
Int. Cl. G06T 11/60 (2006.01); G06T 9/00 (2006.01); G06V 10/44 (2022.01); G06V 40/16 (2022.01)
CPC G06T 11/60 (2013.01) [G06T 9/00 (2013.01); G06V 10/44 (2022.01); G06V 40/171 (2022.01)] 14 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining a random facial image;
inputting the facial image to a semantic extractor trained to extract a feature vector representing one or more semantic features from the facial image;
decomposing the feature vector to extract one or more vectors representing specific features of the facial image;
sampling the one or more vectors representing the specific features;
modifying the extracted feature vector using the sampled vectors; and
generating a new, modified version of the facial image using the modified feature vector;
wherein vectors for identity-invariant features are extracted using a linear regression model; and
wherein vectors for identity-variant features are extracted using a localized semantics learning method.