OpenVX Classifier Extension  9249ea0
Clasifiers Extension


This specification would not be possible without the contributions from this partial list of the following individuals from the Khronos Working Group and the companies that they represented at the time:

  • Frank Brill - Cadence Design Systems
  • Tomer Schwartz - Intel
  • Thierry Lepley - Cadence Design Systems
  • Radha Giduthuri - AMD
  • Jesse Villarreal - TI
  • Victor Eruhimov - Itseez3D
  • Xin Wang - Verisilicon

Background and Terminology

Classification in computer vision is the process of categorizing an image into a finite set of classes or labels. The process normally involves recognition of the dominant content in an image scene. The dominant content should get the strongest confidence score irrespective of the transformation of that content such as scaling, location or rotation.

In this extension we enable the usage of classification methods on an image as a specific class detector. Possible methods can be cascade, SVM, etc. We do not standardize each of these methods, but rather enable their deployment in a standard way. We add to OpenVX a method to import an abstract model: vx_classifier_model. The classifier model can be any kind of classifying technology, and the import API can import any kind of file format. As an example, a vendor can implement in vxImportClassifierModel a parser of the OpenCV cascade XML, and create a cascade classification model similar to the one used in OpenCV.

Kernel names

When using vxGetKernelByName the following are strings specifying the Classifier extension kernel names: