System and method for visual recognition using synthetic training data
AssigneeSynthesis Ai, Inc.
InventorsSergey Nikolenko, Yashar Behzadi
Life of this patent
Estimating the term…
Abstract
A system and method are disclosed for training a model using a training dataset. The training dataset can include only real data, only synthetic data, or any combination of synthetic data and real data. The various aspects of the invention include generation of data that is used to supplement or augment real data. Labels or attributes can be automatically added to the data as it is generated. The data can be generated using seed data. The data can be generated using synthetic data. The data can be generated from any source, including the user's thoughts or memory. Using the training dataset, various domain adaptation models can be trained.
Claims19
Method for training a system, the method comprising:
- generating a plurality of synthetic images, wherein the synthetic images are produced from generative machine learning models, which include neural models;
- combining the plurality of synthetic images with a plurality of real images to produce a training dataset, wherein the training dataset includes synthetic data that is generated based on any one of a plurality of attributes;
- and
- providing the training dataset to the system to train the system to perform visual recognition.
A system for training a model, the system comprising:
- a simulation platform;
- and
- a hybrid neural network in communication with the simulation platform,
- wherein synthetic images are generated by the simulation platform using a generative machine learning model, which includes neural models and deep learning models,
- wherein the synthetic images and real images are inputted to the hybrid neural network, and
- wherein the hybrid neural network generates an enhanced training dataset that is used to further train the model.
A method for improving training datasets, the method comprising:
- generating at least one training dataset, wherein the training dataset includes synthetic images focused on a subject and produced using a generative machine learning model, which includes a deep learning model;
- training a model focused on the subject using the training dataset;
- testing the trained model using seed data;
- generating feedback from the model based on testing;
- and
- improving the training dataset based on the generated feedback.
Dependencies are inferred from the claim text.