How does Low calculate the "repeatability" of his SIFT algorithm?

Question

In his 2004 article, “Distinctive Characteristics of an Image from Scale-Invariant Key Points,” he gave many “repeatability” numbers as a function of XXX, for example, in figures 3,4 and 6, but he did not specify how to calculate “repeatability”.

He actually gave a simple explanation of the "repeatability" in Figure 3 on page 8, which is the "percentage of key points that repeat at the same location and scale in the transformed image."

However, 1) how can we know the weather, the keyword is detected in the converted image, just checking the key points one by one? I don’t think it’s practical, since there are thousands of keyboards.

2) . How close can we see the key point recurring in SAME LOCATION? 3 pixels, 6 pixels? Or not pixels at all?

I don’t know how I can load the paper “Distinctive Image Functions from Scale-Invariant Key Points”. Link here: http://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf

Answer 1

. 9, , . -, " " sqrt(2) , , , . -, " " σ x y, "σ - ( (1) , - )"

, , , . /, , . , (, KLT , , , " "... , , ).

(, ) , . Blerg.