Re: [help-GIFT] Adding text features to Viper/GIFT

[Wolfgang Müller]

Le Mardi 15 Mai 2001 08:51, David Squire a écrit :
> Wolfgang Müller wrote:
>
> [snip]
>
> > For the time being, I guess the best you can do is:
> >
> > 1) slurp in a file containing a list of all your images. Maybe get
> > inspired by url2fts.xml
> > 2) make your feature extractor parse gift-config.mrml
> > 3) add references to the files you generated into gift-config.mrml.
> >
> > Like this, it should be possible for people to add a text index in the
> > same easy fashion as they can add a "normal" gift index.
>
> All this is *way* more complicated than what I have time for now. I am

I see your point. Let's say this is what would be needing for distributing 
it.

> planning to create auxilliary tff (text feature file) files for each text
> file given as input (either on the command line or from a file). These will
> contain (TermID, term frequency) pairs, which will be read by a modified
> version of extract_features after extracting the image features (the
> convention being that we have ImageName.jpg (or gif or whatever)

In this case, you can do much simpler. Do a simple concatenation and change 
the number of features. For outsiders who will read this: In the current 
version of the viper query engine in the GIFT, we are using 

numberOfFeatures: integer

featureId: integer
termFrequency: float 

As a file format. the featureID/termFrequency pair is repeated 
numberOfFeatures times.

So if you want to do a quick hack for adding in your own features:
change numberOfFeatures and simply add your own features in the end.
To do such a concatenation should be *very* short in perl. I think I send it 
here in one previous post. If not it was posted to our internal group list, 
and I repost it here on request.

> accompanied by ImageName.txt, leading to ImageName.tff. The result will be
> fts fi es which the inverted file generator should be able to slurp in
> unmodified - we will just now have another feature group. Conseqently the
> text and image index should look just like a "normal" gift index.

I see. Yes, this works. And if you want to be really safe, you should add 
some items (with the new feature group number) to the 
InvertedFileFeatureDescription.db file.

>
> I realise that this is extremely quick and dirty, but I have a student
> waiting on it who has a tight deadline for the submission if her minor
> thesis. Hopefully I will be able to integrate this into the full framework
> later.
>

I see, I hope this works out :-) and you get this mail in time.

> Cheers,
>
> David
>
> PS. Is it possible to make messages distributed by help-gift have their
> Reply-To address set automatically to address@hidden

It is, but it is strongly discouraged by the FSF. Apparently in the past 
there have been some accidental postings of personal flames to the lists or 
something like that.

Wolfgang