Google Patent Reveals Breakthrough in Synthetic Data Generation for Machine Learning

Google LLC has been granted a patent for an innovative method of generating synthetic data using augmented reality (AR) techniques, which can improve the performance of machine learning models. The patent, filed on November 19, 2019, and published online on October 14, 2025, reveals a system and method for generating synthetic training data using AR techniques, enabling the creation of high-quality synthetic images that closely resemble real-world imagery.

The inventors, Ahmadyan, Adel, Grundmann, Matthias, Hou, Tingbo, and Wei, Jianing, describe a method that involves obtaining a three-dimensional model of a virtual object and data comprising one or more image frames that depict a scene. The system detects one or more planar surfaces in the scene and determines a position and orientation for the virtual object to be in contact with at least one planar surface. The augmented image is generated using an AR rendering engine, and the resulting image is stored as a training image within a machine learning training dataset.

Key Takeaways:

• The patent reveals a system and method for generating synthetic training data using AR techniques, which can improve the performance of machine learning models.
• The method involves obtaining a 3D model of a virtual object and data comprising one or more image frames that depict a scene.
• The system detects one or more planar surfaces in the scene and determines a position and orientation for the virtual object to be in contact with at least one planar surface.
• The augmented image is generated using an AR rendering engine, and the resulting image is stored as a training image within a machine learning training dataset.
• The patent includes 20 claims, which outline the various aspects of the system and method for generating synthetic data.
• The claims cover the process of determining the position and orientation of the virtual object, generating the augmented image, storing the label with the training image, and training a machine-learned model using the synthetic data.
• The patent also covers the use of a random value to determine the position and orientation of the virtual object, as well as the use of a user input to define the position and orientation.

Statistics:

• The patent was filed on November 19, 2019.
• The patent was published online on October 14, 2025.
• The patent includes 20 claims.
• The system and method are capable of generating high-quality synthetic images that closely resemble real-world imagery.
• The synthetic data can be used to train machine learning models that perform better than those trained on real data.

Sources:

• Ahmadyan, Adel. AR-assisted synthetic data generation for training machine learning models. U.S. Patent Number 12444177, filed November 19, 2019, and published online on October 14, 2025.
• Patent URL (for desktop use only): https://ppubs.uspto.gov/pubwebapp/external.html?q=(12444177)&db=USPAT&type=ids