US 12,322,011 B2
Product of variations in image generative models
Yotam Nitzan, Tel-Aviv (IL); Taesung Park, San Francisco, CA (US); Michaël Gharbi, San Francisco, CA (US); Richard Zhang, Burlingame, CA (US); Junyan Zhu, Pittsburgh, PA (US); and Elya Shechtman, Seattle, WA (US)
Assigned to ADOBE INC., San Jose, CA (US)
Filed by ADOBE INC., San Jose, CA (US)
Filed on Nov. 17, 2022, as Appl. No. 18/056,579.
Prior Publication US 2024/0169621 A1, May 23, 2024
Int. Cl. G06T 11/60 (2006.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06T 11/60 (2013.01) [G06V 10/774 (2022.01); G06T 2200/24 (2013.01); G06V 10/82 (2022.01)] 12 Claims
OG exemplary drawing
 
1. A method for image generation, comprising:
obtaining an input image and an attribute value representing an attribute of the input image to be modified;
selecting a basis vector corresponding to the attribute from a set of orthogonal basis vectors of a latent space of an image generation network;
computing a modified latent vector for the input image by applying the attribute value to the selected basis vector; and
generating a modified image by decoding the modified latent vector using a convolutional layer of the image generation network, wherein the modified image comprises a synthesized image that includes the attribute based on the attribute value.