Talk:Scale-invariant feature transform

From Wikipedia, the free encyclopedia
Jump to: navigation, search
WikiProject Robotics (Rated C-class, Mid-importance)
WikiProject icon Scale-invariant feature transform is within the scope of WikiProject Robotics, which aims to build a comprehensive and detailed guide to Robotics on Wikipedia. If you would like to participate, you can choose to edit this article, or visit the project page (Talk), where you can join the project and see a list of open tasks.
C-Class article C  This article has been rated as C-Class on the project's quality scale.
 Mid  This article has been rated as Mid-importance on the project's importance scale.
Note icon
This article has been marked as needing immediate attention.

As this page is directly about SIFT, descriptions of GLOH and SURF should be moved to seperate pages. Only brief mentions and links to GLOH & SIFT should be included on this page.

Clarification of Mathematic Variable Meanings[edit]

I was just wondering if it would be at all possible to have the meanings of the mathematical symbols referenced as well. For someone who is doing research and attempting to learn what all this means it would be extremely helpful. JRSofty (talk) 11:03, 17 January 2008 (UTC)

Image Needed[edit]

Why 0.03?[edit]

This page would really benefit from an explanation why this value is chosen.--bb (talk) 21:52, 8 December 2008 (UTC)

There's no theoretical reason for it to be precisely 0.03. It's there just to filter out features resulting from noise. IMO if you have good data, you could use lower (less restrictive) value. —Preceding unsigned comment added by (talk) 12:02, 10 December 2008 (UTC)


What does it mean that the algorithm is patented? You can't use it in a commercial product? How long will that last? Is this controversial? I'd like to see a section in the article that gives details about this. --Singularitarian (talk) 05:18, 9 January 2009 (UTC)

Generally speaking, it's a bad idea for wikipedia to try to summarize the intellectual property status of ideas. Yes, it has been widely discussed that David Lowe's university got a U.S. Patent based on his work. If you want to know what that means, you need to consult an attorney. It may mean that you can't use the invention in the US; they may also have pending or issued foreign patents. Or they may have let the patent lapse. Or it may be invalid for any of a number of reasons. Or the claims may not apply to the exact algorithm or application that you have. Amateurs like wikipedia writers shouldn't be in the position of trying to assess these things. In the current case, the statement of patent is sourced to some guy's web page, so you don't even know what patent is being referred to. Dicklyon (talk) 05:34, 9 January 2009 (UTC)
While Lowe's work on object recognition is a very valuable contribution to the field of computer vision, a number the ideas and methods used in his work on the SIFT descriptor have also been used by others before his publications or what I would assume his patent application. Tpl (talk) 13:30, 9 January 2009 (UTC)
And that's why it takes a skilled attorney and an expert in the art to go over the claims together and try to figure out what you might or might not want to do if you're worried about infringement. It would be best if wikipedia were silent on all this, in my opinion. Dicklyon (talk) 01:57, 10 January 2009 (UTC)
It may require a very good expert or even require more than one expert to give an appropriate description of prior art ... Tpl (talk) 16:36, 10 January 2009 (UTC)
You might be right, but not all Wikipedia writers are amateurs. Maybe there are patent lawyers out there who have the time ($deity forbid!) to edit such articles and provide the necessary details. Besides, providing information that there is some form of valid IP claim on it would be a good idea, at least it would make sure people are aware they should check first before using this. Icedwater (talk) 06:36, 10 July 2011 (UTC)
Please note that in any case, the patent is not valid outside of the US - the US being the only country crazy/stupid enough to allow for the patenting of algorithms. Meanwhile, the rest of the world actually has some common sense and you're free to use SURF as much as you like. — Preceding unsigned comment added by (talk) 05:53, 14 August 2012 (UTC)

Section "Keypoint descriptor"[edit]

What's the meaning of the following sentence? "This step is image closest in scale to the keypoint's scale." —Preceding unsigned comment added by (talk) 07:48, 1 September 2009 (UTC)

The descriptor is vaguely explained. Lowe's original feature vector is 160-dimensional, not 128. It uses histograms on two adjacent levels in the image pyramid. Clarifications are badly needed, but I'm not a native... —Preceding unsigned comment added by (talk) 06:54, 19 October 2009 (UTC)

Many recent computer vision papers and applications use the SIFT keypoint descriptor without using the other elements in the SIFT framework (e.g. feature point detection, matching method). This section therefore should be more emphasized and described in more details. (talk) 14:05, 19 April 2013 (UTC)

Autopano (Software product)[edit]

I've noticed that searching for AutoPano Pro or AutoPano redirect to this page, but I'm not sure why and I don't think that they should. Unless someone objects I'm going to have a go at creating a page for the AutoPano Pro software product (bear in mind that I am a new user). At the very least it seems like the redirect from AutoPano Pro to this page should be removed, as this page makes no mention to the software product. --Kragen2uk (talk) 06:49, 12 January 2011 (UTC)

I agree there should not be a mere redirect. At least a stub: "AutoPano is a panoramic stitching software using SIFT". You may want to look at the page describing another related software: AutoStitch. If SIFT is a significant technology behind AutoPano, I think it's appropriate to link to it from the SIFT article (there is already a link to AutoStitch). Akinoame (talk) 17:46, 12 January 2011 (UTC)

D( x, y, σ) and D[edit]

The difference between D( x, y, σ) and D in the section 'Interpolation of nearby data for accurate position' is completely unclear to me. I think the article would benefit from an explanation of this. (talk) 13:18, 3 February 2011 (UTC)

Attention needed[edit]

  • Refs - large amounts of text have no refs
  • Content - are all topics covered?
  • Prose and MoS compliance checks
  • Assess on B criteria
  • Article claims "...are easy to match against a (large) database of local features" without a source, even though 128 dimensions are hard to handle for datastructures like kd-trees. I'm new to wikipedia, but just wondering.

Chaosdruid (talk) 02:28, 9 March 2011 (UTC)

Yes, please do work on it. Some good sources about how to use locality-sensitive hashing with SIFT are here. Dicklyon (talk) 00:48, 24 July 2012 (UTC)

Organization: Move Key Stages out of overview?[edit]

I just read through the article in its entirety for the first time on mobile (where the TOC doesn't show), and found it rather hard to follow. The "Key Stages" section seems to dive into way too much detail for an overview, and is confusing when it is presented before a description of the keypoints or descriptors. I suggest the article would be improved if this section was dramatically reduced, and the current contents moved into the algorithm section. Leopd (talk) 17:04, 1 October 2012 (UTC)