help-gift
[Top][All Lists]
Advanced

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

Re: [help-GIFT] Re: Clarification on inverted file


From: MUELLER Henning
Subject: Re: [help-GIFT] Re: Clarification on inverted file
Date: Mon, 03 Sep 2001 16:31:09 +0200

Hi,

sorry that it took me a while to answer the pruning question.

There are two articles on pruning which you might have alreday taken a
look at. 
The feature pruning that is active in the actual version was found to be
the most efficient pruning method that does not change the results very
much. I was experimenting with lossless pruning, but the results were
not much better with our database than without pruning. If you have a
database with groups of very similar images where the differences to
other groups is very high, you might be able to have better results with
lossless pruning.
I experimented with reducing the Scoreboard by the images that would
have no chance anymore to reach the n=20 (for example) top result
images. I also did tests with stopping the feature evaluation when the
gap between the n'th and (n+1)th image was big enough so no change in
the top n would be possible. Both without a big reduction in query time.

The scoreboard pruning and the time-based pruning are untested in the
current GIFT environment but should work as they do not rely on other
parts of the system. They were tested a lot in an older Viper version.

Most of the other pruning questions were already answered in detail by
David and Wolfgang anyways.

MARS actually means Multimedia Archiving and Retrieval Systems (plural).
There is not a single Mars programme but several, it is more like a
project name.

There were several presentation of CBIR systems at ICME 2001 that use
very similar techniques to text retrieval, one even trying to find
visual information that corresponds to words (called keyblocks). It is
by Lei Zhu Chung Tang, Aibing Rao, Aidong Zhang from Buffalo. They use
very very similar techniques to what we do.

Henning
-- 
     ----------------------------------------------------------
     Henning Mueller, Computer Vision Group
     Computer Science Department, University of Geneva
     24, rue du General Dufour, CH-1211 Geneva 4, SWITZERLAND  
     Phone : +41(22)705 7633; fax: +41(22)705 7780
     address@hidden
     ----------------------------------------------------------



reply via email to

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