Device, a method and an apparatus for improving quantization capabilities for computer visual applications
US12175710B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Jun 8, 2022 |
| Grant date | Dec 24, 2024 |
| Priority date | — |
| Expiry date | Jun 20, 2043 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06T2207/20084
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A method and a computational module are provided for carrying out a quantization process of a plurality of channels carrying data received from an image capturing sensor. The computational module comprises: at least one array of processors, configured to a) retrieve data from: a1) a neural network graph, a2) a dataset associated with a data and a3) parameters' values of a neural network model; b) carry out a dynamic range calibration process for the channels received and using the neural network graph for deriving grouping constrains associated with respective channels; c) carry out a grouping optimization based on results obtained for each channel from its respective dynamic range calibration and its grouping constrains; d) arrange the channels so that channels having similar grouping constrains are grouped together into one output channel; and e) calculate required quantization parameters for carrying out a quantization process of the output channels.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.