US 12,067,645 B2
High-resolution controllable face aging with spatially-aware conditional GANs
Julien Despois, Paris (FR); Frederic Flament, Paris (FR); and Matthieu Perrot, Paris (FR)
Assigned to L'Oreal, Paris (FR)
Filed by L'Oreal, Paris (FR)
Filed on Jun. 30, 2021, as Appl. No. 17/363,098.
Claims priority of provisional application 63/046,011, filed on Jun. 30, 2020.
Claims priority of application No. 2009199 (FR), filed on Sep. 11, 2020.
Prior Publication US 2021/0407153 A1, Dec. 30, 2021
Int. Cl. G06T 11/00 (2006.01); A45D 44/00 (2006.01); G06N 3/045 (2023.01); G06N 3/088 (2023.01); G06Q 30/0601 (2023.01); G06T 7/00 (2017.01); G06V 10/44 (2022.01); G06V 10/50 (2022.01); G06V 10/764 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06V 40/16 (2022.01)
CPC G06T 11/00 (2013.01) [A45D 44/005 (2013.01); G06N 3/045 (2023.01); G06N 3/088 (2013.01); G06Q 30/0631 (2013.01); G06Q 30/0643 (2013.01); G06T 7/0012 (2013.01); G06V 10/454 (2022.01); G06V 10/50 (2022.01); G06V 10/764 (2022.01); G06V 10/765 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06V 40/171 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30088 (2013.01); G06T 2207/30201 (2013.01)] 27 Claims
OG exemplary drawing
 
1. A computing device comprising:
a face-effect unit including processing circuitry configured to apply at least one facial effect to a source image and to generate a virtual instance of an applied-effect source image on an interface, the face-effect unit utilizing a deep neural network generator to simulate aging with continuous control over a plurality of age related skin signs between a first image and a translated image of the face, the generator configured to automatically translate the first image using respective aging targets for the skin signs;
wherein the aging targets are provided to the generator as an aging map identifying location zones of the face associated with respective ones of the skin signs, where each zone in the aging map is filled with a respective aging target corresponding to the associated skin sign,
wherein the generator comprises a fully convolutional encoder-decoder comprising residual blocks in the decoder to incorporate the aging targets in the form of aging maps;
wherein the generator is configured using a patch-based training methodology using a portion of a particular training image and a corresponding patch of an associated aging map;
wherein the residual blocks further incorporate location information to indicate a respective location of the portion of the particular training image and the corresponding patch of the associated aging map; and
wherein the location information is provided using respective X and Y coordinate maps defined from a horizontal gradient map and a vertical gradient map related to a height and width (H×W) size of the first image.