Method and electrical device for training cross-domain classifier
US12400429B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Jun 12, 2023 |
| Grant date | Aug 26, 2025 |
| Priority date | — |
| Expiry date | May 17, 2044 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06V40/197
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A method for training a cross-domain classifier includes the following steps: (a) obtaining training samples from a first database and test samples from a second database; (b) performing an inference procedure to the test samples by the classifier to generate corresponding predicted labels; (c) for a certain category, obtaining the training samples and the test samples belonging to this category, and training an generative adversarial network (GAN) according to the obtained training samples and test samples; (d) performing a style conversion to the obtained training samples by the GAN to obtain synthetic samples; (e) merging the synthetic samples with the training samples to train the classifier; and repeating the above steps (b) to (e). The classifier will be suitable for cross-domain databases based on this iterative procedure.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.