help-gift
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [help-GIFT] 'Re: Ups.


From: Wolfgang Müller
Subject: Re: [help-GIFT] 'Re: Ups.
Date: Thu, 13 Apr 2006 17:36:36 +0200
User-agent: KMail/1.9.1

On Thursday 13 April 2006 17:28, David Squire wrote:
> address@hidden wrote:
> > the only reason i went with netPBM was one of my initial goals was to run
> > against a whole one of my images, in native resolution. i've read that
> > PPM has changed formats, and it looked to me like the PPM code in GIFT
> > wouldnt handle any images greater than 256x256.

> GIFT rescales all images to 256x256 before feature extraction. This is
> essential for the way "block features" are extracted and indexed in the
> GIFT. If you have removed this rescaling, then you will have broken the
> feature extraction, since feature IDs depend on block positions, that
> rely on a fixed input image size.

Yes and no. If she makes the quad tree one level deeper, there is no problem 
with this. The assignment of features to feature groups is generated by the 
same program she is running for extracting features from each image. She does 
get problems, though, if she does not generate the proper feature information 
file for her new features.

> I am not arguing that this is a particularly sensible thing to do, but
> that is the way the GIFT works at present.
>
> The PPM code with the GIFT happily handles images of any size - I have
> used that PPM library in several other applications.

This is good to know.

> It is the GIFT 
> feature extraction that *requires* the rescaling to 256x256.

Yes, I remember the discussion. Many things the GIFT does are due to decisions 
that seemed reasonable at the time. Makes me think of this article you quoted 
at the time that was about the lifetime of engineering decisions.

Cheers,
Wolfgang

-- 
Dr. Wolfgang Müller
LS Medieninformatik
Universität Bamberg
Check out the SIG MM web site http://www.sigmm.org




reply via email to

[Prev in Thread] Current Thread [Next in Thread]