Experimental Evaluation of Multiplicative Kernel SVM Classifiers for Multi-Class Detection

Abstract

We consider the multi-class object detection approach based on a non-parametric multiplicative kernel, which provides both separation against backgrounds and feature sharing among foreground classes. The training is carried out through the SVM framework. According to the obtained support vectors, a set of linear detectors is constructed by plugging the foreground training samples into the multiplicative kernel. However, evaluating the complete set would be inefficient at runtime, which means that the number of detectors has to be reduced somehow. We propose to reduce that number in a novel way, by an appropriate detector selection procedure. The proposed detection approach has been evaluated on the Belgian traffic sign dataset. The experiments show that detector selection succeeds to reduce the number of detectors to the half of the number of object classes. We compare the obtained performance to the results of other detection approaches and discuss the properties of our approach.

Files

Full Paper as PDF

BibTeX Citation

DOI

10.20532/ccvw.2014.0017

https://doi.org/10.20532/ccvw.2014.0017

BibTeX

@InProceedings{10.20532/ccvw.2014.0017,
  author =       {Valentina Zadrija and Sini{\v s}a {\v S}egvi{\' c}},
  title =        {Experimental Evaluation of Multiplicative Kernel SVM
                  Classifiers for Multi-Class Detection},
  booktitle =    {Proceedings of the Croatian Compter Vision Workshop,
                  Year 2},
  pages =        {50-55},
  year =         2014,
  editor =       {Lon{\v c}ari{\' c}, Sven and Suba{\v s}i{\' c},
                  Marko},
  address =      {Zagreb},
  month =        {September},
  organization = {Center of Excellence for Computer Vision},
  publisher =    {University of Zagreb},
  abstract =     {We consider the multi-class object detection
                  approach based on a non-parametric multiplicative
                  kernel, which provides both separation against
                  backgrounds and feature sharing among foreground
                  classes. The training is carried out through the SVM
                  framework. According to the obtained support
                  vectors, a set of linear detectors is constructed by
                  plugging the foreground training samples into the
                  multiplicative kernel. However, evaluating the
                  complete set would be inefficient at runtime, which
                  means that the number of detectors has to be reduced
                  somehow. We propose to reduce that number in a novel
                  way, by an appropriate detector selection
                  procedure. The proposed detection approach has been
                  evaluated on the Belgian traffic sign dataset. The
                  experiments show that detector selection succeeds to
                  reduce the number of detectors to the half of the
                  number of object classes. We compare the obtained
                  performance to the results of other detection
                  approaches and discuss the properties of our
                  approach.},
  doi =          {10.20532/ccvw.2014.0017},
  url =          {https://doi.org/10.20532/ccvw.2014.0017}
}