This forum is disabled, please visit https://forum.opencv.org
 | 1 | initial version |
You should read the original SIFT paper first. The size specifies the layer in the gaussian pyramid (octave & sub-layers) in which the corner-like feature recieves the highest value in the response-function. Also, the extracted gradients in the neighborhood of the feature lead to the direction of the detected keypoint.
As you can see, it is very hard to guess some "handwritten values". Maybe you try to use DSIFT or some kind of other feature detector for your project?
