Wave-based detector

This page provides access to the source code of our WaDe keypoint detector in the form of precompiled binaries and MATLAB mex files. WaDe is an implementation of our work on a novel way to define scale-covariant keypoints, by exploiting the properties of the wave equation. You can find more details in our CVPR 2013 paper [1]. Scissors

The images show an example of WaDe keypoints found on different kind of objects in diverse datasets.

UntexturedModel1 Strongest 300 keypoints Strongest 300 Keypoints

You can download a ZIP file with the binaries and instructions from here.
The binaries are free to use for research purposes. Please refer to the command line help and the README file for instructions on how to use it.

The datasets used to produce the results in [1] are

  • Oxford dataset, available here
  • DTU Robot dataset, available here
  • An untextured objects dataset, that we provide here

Version history

  • 08/10/2013 Added a command-line flag to export the keypoint sharpness
  • 16/09/2013 Static C++ run-time linking in Linux binary (thanks to Nicola Fioraio)
  • 25/07/2013 Removed dependency from CUDA in Linux binary (thanks to Karel Lenc)
  • 05/07/2013 First release

References

  1. S. Salti, A. Lanza, L. Di Stefano, “Keypoints from Symmetries by Wave Propagation”, IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2013.