A Fast Local Descriptor for Dense Matching


Authors: Engin Tola, Vincent Lepetit, Pascal Fua


Q1: How depth estimation is related with object recognition?

Objects are located in a 3D environment and in order to recognize them correctly, we need to be able to recreate their layout in a scene. With such an aid, we cn successfully determine the object boundaries.

Q2: What does the descriptor contain?

It's a concatenation of vectors. The first vector is the Gaussian of the center point with a \Sigma\~0~, the second set of vectors are circles lying on circle R\~1~, the third set of vectors are circles lying of circle R\~2~... Each vector contains orientation maps after a Gaussian convolution.

Q3: On which datasets did the authors try the technique?

As fas as I can get, it's a custom dataset that contains the view of the same scene from many perspectives.

Q4: What's the salience criteria for keypoints?

The aim of the technique is not matching these keypoints to each other by selecting the most appropriate ones. The computation is done on all keypoints. Hence no criteria for keypoint filtering is reported.