SIFT descriptors values: OpenCV vs VLFeat

asked 2016-12-08 05:52:15 -0500

lovaj gravatar image

I'm trying to compare the SIFT implementation of OpenCV and VLFeat.

I noticed that the descriptors value for VLFeat are integers, such as:

0 0 0 0 0 0 0 0 0 0 0 17 45 20 26 0 1 ..

While for OpenCV:

0.0391555 0 0 0.0998274 0.235747 0 0 0.0276871 0.156622 ...

Notice that these are descriptors for 2 different images.

I have two questions:

  1. Why they have two different values?
  2. If I'm going to need the OpenCV representation for k-means using VLFeat (and then VLAD encoding) do I need to change these values?
edit retag flag offensive close merge delete

Comments

  1. opencv's sift descriptors are normalized
berak gravatar imageberak ( 2016-12-08 07:29:54 -0500 )edit

And what are the consequences of that? There advantages/disadvantages of a such representation?

lovaj gravatar imagelovaj ( 2016-12-08 07:52:43 -0500 )edit