I'm trying to produce [VLAD][1] codes starting from SIFT descriptors using OpenCV implementation with the [Oxford Building Dataset][2].
SIFT has many parameters such as #octaves, #scales, contrastThreshold and edgeThreshold. The right values for these parameters can significantly improve descriptors VLAD precision, so tuning is needed.
What is a good way for this task? Any particular procedure is present? Any suggestion is more than welcome!
[1]: http://www.vlfeat.org/api/vlad-fundamentals.html
[2]: http://www.robots.ox.ac.uk/~vgg/data/oxbuildings/
↧