US 11,537,813 B1Granted

System for synthesizing data

AssigneeAmazon Technologies, Inc.

InventorsIgor Kviatkovsky, Nadav Israel Bhonker, Alon Shoshan, Manoj Aggarwal, Gerard Guy Medioni

Life of this patent

Estimating the term…

Loading patent…

Abstract

During a training phase, a first machine learning system is trained using actual data, such as multimodal images of a hand, to generate synthetic image data. During training, the first system determines latent vector spaces associated with identity, appearance, and so forth. During a generation phase, latent vectors from the latent vector spaces are generated and used as input to the first machine learning system to generate candidate synthetic image data. The candidate image data is assessed to determine suitability for inclusion into a set of synthetic image data that may be used for subsequent use in training a second machine learning system to recognize an identity of a hand presented by a user. For example, the candidate synthetic image data is compared to previously generated synthetic image data to avoid duplicative synthetic identities. The second machine learning system is then trained using the approved candidate synthetic image data.

Claims20

Dependent claims

A system comprising:

a memory, storing first computer-executable instructions;
and
one or more hardware processors to execute the first computer-executable instructions to:
acquire a first set of actual images of a portion of a hand, wherein each image is representative of an appearance of the hand and identifying features of the hand;
train a first machine learning system using the first set of actual images;
determine, based on the trained first machine learning system:
a first latent vector space that is associated with appearance of the hand;
and
a second latent vector space that is associated with identifying features of the hand;
determine a first synthetic vector from the first latent vector space;
determine a second synthetic vector from the second latent vector space;
generate candidate synthetic image data using the first synthetic vector and the second synthetic vector as inputs to the trained first machine learning system, wherein the candidate synthetic image data is associated with a first identity;
determine the candidate synthetic image data is suitable for training;
and
train a second machine learning system using input data comprising the candidate synthetic image data.

12 more elements

A method comprising:

acquiring a first set of images, wherein each image is representative of at least a portion of a person;
training a first machine learning system using the first set of images;
determining, based on the trained first machine learning system:
a first latent vector space that is associated with appearance of a person;
and
a second latent vector space that is associated with identifying features of the person;
determining a first synthetic vector from the first latent vector space;
determining a second synthetic vector from the second latent vector space;
generating candidate synthetic image data using the first synthetic vector and the second synthetic vector as inputs to the trained first machine learning system, wherein the candidate synthetic image data is associated with a first identity;
determining the candidate synthetic image data is suitable for use as input data;
and
processing the input data using a second machine learning system using the input data comprising the candidate synthetic image data.

9 more elements

A system comprising:

a memory, storing first computer-executable instructions;
and
one or more hardware processors to execute the first computer-executable instructions to:
acquire a first set of images, wherein each image is representative of at least a portion of a person;
train a first machine learning system using the first set of images;
determine, based on the trained first machine learning system, one or more latent vector spaces;
determine one or more synthetic vectors from the one or more latent vector spaces;
generate candidate synthetic image data using the one or more synthetic vectors, wherein the candidate synthetic image data is associated with a first identity;
determine the candidate synthetic image data is suitable for use as input data;
and
process at least a portion of the input data comprising the candidate synthetic image data using a second machine learning system.

8 more elements

Dependencies are inferred from the claim text.