Re: GIFT query features

Hi,

> Hi,
> > <mrml ...>
> > ...
> > <configure-session> ...
> >        cui-block-color-histogram="true"
> >        cui-block-color-blocks="false"
> >        cui-block-texture-histogram="true"
> >        cui-block-texture-blocks="true"
> > </configure-session>
> > ...
> > </mrml>
> >
> > GIFT will apply only the colour blocks and nothing else? 
> 
> Yes, but cui-block... are attributes of the algorithm tag (you will see when 
> looking closely at the gift-log.mrml). Also make sure, that it really uses 
> session and algorithm you want it to use. One way to make sure would be to 
> base your algorithm on adefault. Then the maximum score you can achieve is 
> 0.25 if you're using one out of four feature groups.

Yup. Thanks for that.

> 
> > This'd essentially 
> > be equivalent to turning the rest of the image features off had I used a
> > GUI interface. Can I also assume that it'll be the same for the other
> > features?
> 
> Yes.

I suspect that what I'm doing above isn't quite the same as disabling the
other features in a GUI interface though. 

Here's my xml query using only colour blocks:

         <algorithm
         algorithm-id="a-cidf"
         algorithm-type="adefault"
         collection-id="c-37-46-17-21-6-105-4-201-0"
         cui-block-color-histogram="true"
         cui-block-color-blocks="false"
         cui-block-texture-histogram="true"
         cui-block-texture-blocks="true"
         cui-pr-percentage-of-features="70" >
         </algorithm>

Here's the xml query that's generated by GIFT's GUI interface. I disabled  
all the features except for colour blocks:

         <algorithm 
         algorithm-id="a-cidf" 
         algorithm-type="a-cidf"
         collection-id="c-37-46-17-21-6-105-4-201-0"
         cui-pr-percentage-of-features="70"
         cui-block-color-blocks="false"
         cui-block-texture-blocks="true"
         cui-block-texture-histogram="true"
         cui-block-color-histogram="true" >
	 </algorithm>

With the latter query, the similarity scores exceed 0.25, which tells me that
it's using more than just one out of four features. However, if I changed the
"algorithm-type" to "a-cidf" for the former query, I would then get the the
same similarity scores. Can you please shed some light?

Thanks,
Jun Jie