| CPC G06T 13/40 (2013.01) [G06N 3/0475 (2023.01); G06T 19/006 (2013.01); G06T 19/20 (2013.01); G06T 2219/2012 (2013.01); G06T 2219/2021 (2013.01)] | 18 Claims |
|
1. A method comprising:
accessing, by one or more processors of a device, a plurality of images that depict one or more persons;
receiving input that identifies a given person of the one or more persons who is depicted in an individual image of the plurality of images;
extracting features of the given person depicted in the individual image;
applying a machine learning model comprising a generative artificial neural network (GAN) to the extracted features of the given person to generate an avatar that resembles the given person, the machine learning model trained by performing training operations comprising:
receiving training data comprising a plurality of training images that depict a plurality of training persons and corresponding ground-truth avatars that resemble the training persons;
applying the machine learning model to a first training image of the plurality of training images to generate an estimated avatar corresponding to a given training person depicted in the first training image;
retrieving a ground-truth avatar that resembles the given training person;
computing a deviation between the ground-truth avatar that resembles the given training person and the estimated avatar corresponding to the given training person depicted in the first training image; and
updating one or more parameters of the machine learning model based on the computed deviation; and
applying one or more augmented reality (AR) fashion items to the avatar to generate an image that resembles the given person wearing the one or more AR fashion items.
|