[help-GIFT] Re: Clarification on inverted file

[David Squire]

suryani lim wrote:

> I am trying out GIFT at the moment, to see if I can use it and to test
> your published algorithms on my database. I had some problems indexing
> my images when I first used it. It's looking good now but I am unsure
> which combination of algorithms are implemented in Classical IDF. You
> have published several weight functions and pruning algorithms.
>
> For pruning, it seems that only sorted weight and lossy pruning is
> implemented, with the pruning features set at 70%? Is that right?

It would be best to post questions like this to the help-gift mailing list (see 
http://www.gnu.org/software/gift/gift.html#mailinglist - I've copied this to 
it). Wolfgang should know exactly which weighting function is implemented in 
the present version (it probably includes Rocchio stuff, and separate 
normalization), and Henning and/or Wolfgang should know about the pruning.

> > PS. Have you read our papers on improving response time
> > through search pruning with inverted files? Check
> > http://www.csse.monash.edu.au/~davids/publications/, under 1999.
>
> Yes, I sure have. I am very glad that you have your
> publications on-line. I have also been searching who else might be
> working on Inverted File for CBIR. Looks like your team is the only
> one who has published any work on it.

Few people say that they are using inverted files, but quite a few techniques 
use the same idea, which is to index images by features, rather than vice 
versa. Smith and Chang's colour sets is a good example (see 
http://www.ctr.columbia.edu/~jrsmith/html/pubs/PAMI/pami_final_1.html). There 
are other examples in the literature; sometimes the term "hashing" is used.

Cheers,

David


--
Dr. David McG. Squire, Postgraduate Research Coordinator (Caulfield)
Computer Science and Software Engineering, Monash University, Australia
http://www.csse.monash.edu.au/~davids/   http://viper.unige.ch/
Do/Don't want HTML mail? Let me know.