Crowd-sourced data collection and labelling using gaming mechanics for machine learning model training
US11712633B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Mar 12, 2020 |
| Grant date | Aug 1, 2023 |
| Priority date | — |
| Expiry date | Nov 4, 2041 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06V20/68
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A gamified application is provided for users to feed animated virtual characters with images of real-world food items. The images fed to the virtual characters are to be uploaded to a data store in a cloud environment, for use in training a custom machine learning model. A server in the cloud environment receives a photo of a food item fed to a virtual character in an augmented reality environment in the gamified application executing on a user device, invokes the custom machine learning model to generate classification information for the photo, sends the classification information to the user device for verification by a user, and stores the verified information to the data store used for periodically training the machine learning model. Over time, the data store would include a large volume of food images with label data verified by a large number of users.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.