Non-adversarial image generation using transfer learning
AssigneeAdobe Inc.
InventorsPuneet Mangla, Balaji Krishnamurthy
Life of this patent
Estimating the term…
Abstract
In implementations of systems for non-adversarial image generation using transfer learning, a computing device implements a generation system to receive input data describing random noise. The generation system generates a latent representation in a latent space of a machine learning model based on the random noise using a transformer model that is trained to generate latent representations in the latent space. A digital image is generated using the machine learning model based on the latent representation that depicts an object that is visually similar to objects depicted in digital images of a training dataset used to train the machine learning model based on a perceptual loss.
Claims20
A method comprising:
- receiving, by a computing device, input data describing random noise;
- generating, by the computing device, a latent representation in a latent space of a machine learning model based on the random noise using a transformer model that is trained at least in part based on interpolation of latent representations of digital images to generate latent representations in the latent space;
- and
- generating, by the computing device, a digital image using the machine learning model based on the latent representation, the digital image depicting an object that is visually similar to objects depicted in digital images of a training dataset used to train the machine learning model based on a perceptual loss.
A system comprising:
- a memory component;
- and
- a processing device coupled to the memory component, the processing device to perform operations comprising:
- receiving training data describing a set of digital images;
- training a machine learning model using the training data based on a perceptual loss;
- training a transformer model based on latent representations of digital images included in the set of digital images to generate latent representations in a latent space of the machine learning model, the training including use of interpolations of the latent representations of the digital images;
- and
- generating a digital image depicting an object that is visually similar to objects depicted in the digital images included in the set of digital images using the transformer model and the machine learning model based on random noise.
A non-transitory computer-readable storage medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:
- receiving input data describing random noise;
- generating a latent representation in a latent space of a machine learning model based on the random noise using a transformer model that is trained based on latent representations of digital images and interpolations of the latent representations of the digital images to generate latent representations in the latent space;
- and
- generating a digital image using the machine learning model based on the latent representation, the digital image depicting an object that is visually similar to objects depicted in digital images of a training dataset used to train the machine learning model based on a perceptual loss.
Dependencies are inferred from the claim text.