Video training data for machine learning detection/recognition of products
Assigneesalesforce.com, inc.
InventorAlex Papli
Life of this patent
Estimating the term…
Abstract
Described herein are systems, apparatus, methods and computer program products configured for image detection/recognition of products. The disclosed systems and techniques utilize video data to provide the necessary number of images and view angles needed to train a machine learning product detection/recognition system to recognize a specific product within later provided images. In various embodiments, a user may provide video data and the video data may be transformed in a manner that may aid in training of the machine learning system.
Claims20
A computer-implemented method implemented in a machine learning environment, the method comprising:
- creating training data by modifying video data to highlight a first object within a plurality of frames, the modifying comprising:
- identifying the first object within a first plurality of frames;
- determining non-first object pixels within the first plurality of frames;
- creating first modified video data by replacing, independent of the pixels of the first object, the non-first object pixels with first background pixels in the first plurality of frames;
- and
- creating second modified video data by replacing, independent of the pixels of the first object, the first background pixels with second background pixels in the first plurality of frames;
- and
- providing the training data to a machine learning program, wherein the training data is configured to train the machine learning program to identify the first object within an image.
A computer program product comprising computer-readable program code capable of being executed by one or more processors in an object detection/recognition environment when retrieved from a non-transitory computer-readable medium, the program code comprising instructions configurable to cause operations comprising:
- creating training data by modifying video data to highlight a first object within a plurality of frames, the modifying comprising:
- identifying the first object within a first plurality of frames;
- determining non-first object pixels within the first plurality of frames;
- creating first modified video data by replacing, independent of the pixels of the first object, the non-first object pixels with first background pixels in the first plurality of frames;
- and
- creating second modified video data by replacing, independent of the pixels of the first object, the first background pixels with second background pixels in the first plurality of frames;
- and
- providing the training data to a machine learning program, wherein the training data is configured to train the machine learning program to identify the first object within an image.
A computing system implemented using a server system implemented in an object detection/recognition environment, the computer system configurable to cause execution of operations comprising:
- creating training data by modifying the video data to highlight a first object within a plurality of frames, the modifying comprising:
- identifying the first object within a first plurality of frames;
- determining non-first object pixels within the first plurality of frames;
- creating first modified video data by replacing, independent of the pixels of the first object, the non-first object pixels with first background pixels in the first plurality of frames;
- and
- creating second modified video data by replacing, independent of the pixels of the first object, the first background pixels with second background pixels in the first plurality of frames;
- and
- providing the training data to a machine learning program, wherein the training data is configured to train the machine learning program to identify the first object within an image.
Dependencies are inferred from the claim text.