When I run sift.compute (or sift.detectAndCompute), in python3 (via anaconda), the descriptors get truncated to integer arrays instead of floating point.
I am wondering how do I get the descriptors in python as the original floating point numbers?
SIFT descriptors ARE float, come rain or come shine:
>>> sift = cv2.xfeatures2d.SIFT_create()
>>> im = cv2.imread("img/crack.jpg",0)
>>> im.shape
(379, 500)
>>> kp,des = sift.detectAndCompute(im,None)
>>> des
array([[ 47., 9., 6., ..., 0., 0., 2.],
[ 8., 3., 5., ..., 0., 7., 15.],
[ 28., 14., 2., ..., 4., 2., 1.],
...,
[ 88., 6., 1., ..., 21., 17., 5.],
[ 66., 5., 1., ..., 23., 22., 7.],
[ 2., 0., 0., ..., 2., 1., 1.]], dtype=float32)
(those are actually histogram bins, counters, they are integer numbers (in a float array), not truncated at all)
Asked: 2018-01-02 04:25:16 -0600
Seen: 1,072 times
Last updated: Jan 02 '18
" the descriptors come as np.uint8 arrays." -- not so, here. please show, how you arrrive at that.
I run
sift = cv2.xfeatures2d.SIFT_create()and then callkp, desc = sift.detectAndCompute()and I get the descriptors. If it helps, I am using OpenCV 3.1.0.@berak I see what you mean. I have updated my question. Sorry for the confusion.