mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-30 22:40:17 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Updated Graph API (markdown)

Vadim Pisarevsky 2018-10-16 14:52:45 +03:00
parent 2c7e210da6
commit ca02380dd1
1 changed files with 14 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
Graph-API.md

@ -1,7 +1,14 @@
Graph API module
- OpenCV 4.0 comes with a new experimental Graph API module (see opencv/modules/gapi). This is a new experimental API which allows to enable offload and optimizations for image processing / CV algorithms on pipeline level.
- The idea behind G-API is to declare image processing task in form of expressions and then submit it for execution – using a number of available backends. At the moment, there’s reference “CPU” (OpenCV-based) and experimental “Fluid’ backends available, with OpenCL (GPU) and other backends coming up next.
- G-API is an uncommon OpenCV module since it acts as a framework: it provides means for declaring operations, building graphs of operations, and finally implementing the operations for a particular backend. G-API model enforces separation between interfaces and implementations, so once an algorithm is expressed in G-API terms, it can be scaled/ported/offloaded to a new platform easily.
- G-API CPU (OpenCV) backend implements G-API standard functions using OpenCV itself (core/imgproc modules) and acts as a quick prototyping/porting/testing backend. If you have an image processing algorithm composed of OpenCV-like functions already, you would be able to switch quickly to G-API by using this backend.
- G-API Fluid backend implements a cache-efficient execution model and allows to save memory dramatically – e.g. a 1.5GB image processing pipeline fits into 750KB memory footprint with G-API/Fluid. G-API comes with a number of operations implemented for Fluid backend, so one can switch OpenCV/Fluid operations within a graph easily and even mix both in the same graph.
- Samples and documentation are WIP and will be completed during 4.0 release timeframe.
What it is?
===========
* OpenCV 4.0 comes with a new experimental Graph API module (see opencv/modules/gapi). This is a new experimental API which allows to enable offload and optimizations for image processing / CV algorithms on pipeline level.
* The idea behind G-API is to declare image processing task in form of expressions and then submit it for execution – using a number of available backends. At the moment, there’s reference “CPU” (OpenCV-based) and experimental “Fluid’ backends available, with OpenCL (GPU) and other backends coming up next.
* G-API is an uncommon OpenCV module since it acts as a framework: it provides means for declaring operations, building graphs of operations, and finally implementing the operations for a particular backend. G-API model enforces separation between interfaces and implementations, so once an algorithm is expressed in G-API terms, it can be scaled/ported/offloaded to a new platform easily.
* G-API CPU (OpenCV) backend implements G-API standard functions using OpenCV itself (core/imgproc modules) and acts as a quick prototyping/porting/testing backend. If you have an image processing algorithm composed of OpenCV-like functions already, you would be able to switch quickly to G-API by using this backend.
* G-API Fluid backend implements a cache-efficient execution model and allows to save memory dramatically – e.g. a 1.5GB image processing pipeline fits into 750KB memory footprint with G-API/Fluid. G-API comes with a number of operations implemented for Fluid backend, so one can switch OpenCV/Fluid operations within a graph easily and even mix both in the same graph.
* Samples and documentation are WIP and will be completed during 4.0 release timeframe.