face and audio and ...

[Paul Over <over@nist.gov>]

Hi,

The minimum size requirement for the face feature was included
as a concession to systems who need some minimal number of pixels 
to work with. Given the questions/comments on this limit in the
feature definition, I think it will be best to drop the 1/4 minimum.

As with the other features, if it is recognizable by a human
then we will consider it an appropriate target for a system - even
if it is beyond current system capabilties. 

Systems will do the best they can. Some will attempt to recognize
faces smaller than 1/4 of the frame etc.; others won't. If some 
can recognize smaller faces - ones large enough for the assessors 
to recognize them as faces by the feature definition - then they 
should be rewarded.

As for audio, the likely mismatch with shots has been noted ,but
for this year we will accept that. If some music or speech spans
multiple shots then the feature will be true for those multiple
shots.

---

Let's concentrate now on resolving the master shot reference 
questions. They affect both the feature extraction and the search 
tasks. Georges has kindly provided us with information on the 
distribution as well as some alternative ways forward.

- Paul
-- 
Paul Over - Retrieval Group
	    Information Access Division
	    Information Technology Laboratory
	    National Institute of Standards and Technology
	    Bldg. 225  Rm. A211  (Mailstop 8940)
	    Gaithersburg, MD  20899-8940   USA
	    Voice: 301 975-6784    Fax: 301 975-5287