Maximum entropy

[Emeline]

Hi,

I'm trying to use maximum entropy in FIRE and I have a few questions
about it:
1) Which framework is used to build the model with training data?
2) To train the model, predicates and outcomes are needed but I don't
really understand how we deduce them from the following features :
fi(Q,X) := di(Qi,Xi).

Thanks in advance,

Emeline

[Thomas Deselaers]

Hi Emeline, 

We have used an internal toolkit for that. 

To use this, you will need to change some code to make use of models trained with another maxent toolkit. 

One option would be yasmet

http://www-i6.informatik.rwth-aachen.de/web/Software/YASMET.html

another one could be  MEGAM

http://www.cs.utah.edu/~hal/megam/

but in any case you will probably have to write some wrapper code. 

Regrading the features: 

We just used the distances as continuous feature functions in maxent. The outcomes are whether an image is relevant or not. 

So, say you have 4 different image descriptors. This will give you one distance per descriptor, so you will have 4 dimensional feature vectors and typically you have very many negative samples (all non-relevant images) and only a few positive samples (all relevant images) per query image.

Best,thomas

--
You received this message because you are subscribed to the Google Groups "FIRE - Flexible Image Retrieval Engine" group.
To post to this group, send email to fire...@googlegroups.com.
To unsubscribe from this group, send email to fire-cbir+...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/fire-cbir?hl=en.

--
http://thomas.deselaers.de