Transformer-based multi-scale pedestrian re-identification method
US12056950B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Oct 24, 2023 |
| Grant date | Aug 6, 2024 |
| Priority date | — |
| Expiry date | Oct 24, 2043 |
Classification
- Technology area (CPC Y)Emerging Cross-Sectional Technologies
- CPC primaryY02T10/40
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
The invention discloses a Transformer-based multi-scale pedestrian re-identification method. The present invention proposes a pedestrian re-identification network based on multi-scale pedestrian feature extraction and Transformer. Firstly, we designed a multi-scale feature cascade module, which aims to mine detailed feature information of pedestrians at different depths and scales, so as to obtain stronger feature representation. Secondly, we constructed a feature extraction based on Transformer to learn pedestrian features at a global scale. Finally, the features output by the Transformer are aggregated to obtain a better expression of pedestrian features, thereby improving the discrimination ability of the model. The result shows that this method has better robustness and adaptive ability, and effectively enhances the generalization ability of the model.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.