Systems and methods for creating training data
AssigneePlus One Robotics, Inc.
InventorsJonathan Lwowski, Abhijit Majumdar
Life of this patent
Estimating the term…
Abstract
Training images can be synthesized in order to obtain enough data to train a model (e.g., a neural network) to recognize various classifications of a type of object. Images can be synthesized by blending images of objects labeled using those classifications into selected background images. To improve results, one or more operations are performed to determine whether the synthesized images can still be used as training data, such as by verifying one or more objects of interested represented in those images is not occluded, or at least satisfies a threshold level of acceptance. The training images can be used with real world images to train the model.
Claims20
A computing system for generating training data in an automated robotic package conveyance system, wherein the generated training data may be used to identify packages and/or package properties in a robotic conveyance system, the computing system comprising:
- a computing device processor;
- and
- a memory device including instructions that, when executed by the computing device processor, enables the computing system to:
- determine a set of classifications corresponding to a type of object;
- obtain a set of catalog images including representations of objects having labels corresponding to the set of classifications;
- obtain foreground imaging data including representations of objects having labels corresponding to the set of classifications and being in real-world environments;
- obtain background imaging data including representations of at least some of the real-world environments;
- identify portions of the foreground imaging data including the representations of the objects;
- identify portions of the background imaging data including representations of the real-world environments;
- blend the portions of the foreground imaging data and the portions of the background imaging data to generate a set of composite training images;
- analyze the set of composite training images to determine a subset of composite training images, individual representations of the objects in the subset of composite training images satisfying acceptance criteria;
- and
- provide the subset of composite training images and the set of catalog images for use in training a model for the type of object.
A computer-implemented method for generating training data in an automated robotic package conveyance system, wherein the generated training data may be used to identify packages and/or package properties in a robotic conveyance system, the computer-implemented method comprising:
- determining a set of classifications corresponding to a type of object;
- obtaining a set of catalog images including representations of objects having labels corresponding to the set of classifications;
- obtaining foreground imaging data including representations of objects having labels corresponding to the set of classifications and being in real-world environments;
- obtaining background imaging data including representations of at least some of the real-world environments;
- identifying portions of the foreground imaging data including the representations of the objects;
- identifying portions of the background imaging data including representations of the real-world environments;
- blending the portions of the foreground imaging data and the portions of the background imaging data to generate a set of composite training images;
- analyzing the set of composite training images to determine a subset of composite training images, individual representations of the objects in the subset of composite training images satisfying acceptance criteria;
- and
- providing the subset of composite training images and the set of catalog images for use in training a model for the type of object.
A non-transitory computer readable storage medium storing instructions for generating training data in an automated robotic package conveyance system, wherein the generated training data may be used to identify packages and/or package properties in a robotic conveyance system, the instructions when executed by at least one processor of a computing system, causes the computing system to:
- determine a set of classifications corresponding to a type of object;
- obtain a set of catalog images including representations of objects having labels corresponding to the set of classifications;
- obtain foreground imaging data including representations of objects having labels corresponding to the set of classifications and being in real-world environments;
- obtain background imaging data including representations of at least some of the real-world environments;
- identify portions of the foreground imaging data including the representations of the objects;
- identify portions of the background imaging data including representations of the real-world environments;
- blend the portions of the foreground imaging data and the portions of the background imaging data to generate a set of composite training images;
- analyze the set of composite training images to determine a subset of composite training images, individual representations of the objects in the subset of composite training images satisfying acceptance criteria;
- and
- provide the subset of composite training images and the set of catalog images for use in training a model for the type of object.
Dependencies are inferred from the claim text.