|
Re: Annotation
|
Clement Leung
|
Aug 14, 2001 21:22 PDT
|
Giordano
I agree that the nouns should be singular (also in view of Corinne's
comments), but the system should be able to match both.
I also agree that the non-visual attributes should be disabled, otherwise
we may have to deal with unnecessary generality/complications (subjective
interpretations?), and that may cloud the issue of visual information
retrieval.
-- Clement
| | Would it be acceptable to establish a convention that all nouns have to
be singular, meaning the image contains at least one object of the
specified category?
A while back Corinne sent out the structured annotation types based on
the MPEG templates, see
http://www.benchathlon.net/sas.html
How about having an index field for each of the seven types, instead of
fitting everything in a single thesaurus?
Also, since to build the ground truth we only need the visual
attributes, should we disable the non-visual attributes for this
annotation round? Each entry in the thesaurus would then have a boolean
field indicating if it is visual.
Giordano
http://www.benchathlon.net/
|
|
|
 |
|
