On page 25 of David Lowe’s paper, Distinguishing Image Features from Scale-Invariant Key Points, in 2004, he states: “Their calculation is efficient so that several thousand key points can be extracted from a typical image with almost real performance on standard PC hardware.” Here is the link: http://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf
However, I tested the SIFT algorithm using Andrea Vedaldi sift ++ (aka VLFeat), which is a C ++ implementation, on a 640x480 image, and it cost 0.839 seconds to extract about 3000 key points from a single image. My computer is an Intel i7 2600k with 16 GB of RAM. Here is the code link: http://www.vlfeat.org/~vedaldi/code/siftpp.html
Honestly, I think it’s rather strange if SIFT can achieve real-time speed, since it needs to extract as many key points from one image.
Does anyone know how fast SIFT is on a modern PC?
source
share