TRECVID - CWI response

["Arjen P. de Vries" <Arjen.de.Vries@cwi.nl>]

>1) MPEG-1:  At the workshop at TREC 2001 several groups indicated they
>   strongly preferred that the data be in MPEG-1, even if this meant 
>   transcoding from e.g., MPEG-2.
> 
> ===> Which groups feel this way?
> 
> ===> Does anyone object to an MPEG-1 only test collection? (Possibly with
>      MPEG-2 available for display to humans.)

We indicated to prefer MPEG-1 last year, however at this point it 
shouldn't really be a problem to handle MPEG-2 too by transcoding to
MPEG-1. Note that at the archive about 1/3 of the data is already available 
in MPEG-1 from the archive. 
In any case, if the dataset would contain MPEG-2 data, it should not 
really be a problem as (free) transcoding tools are available 
enabling anybody to transcode the data locally, if needed.

>2) Common shot definition:
>
>===> Does any one object strongly to using a common set of shot boundaries
>     for the search task?
>
>===> Which groups would volunteer to submit such a set for the total
>     search collection?  NIST would blend these to form the final common set.

If (and only if) the 'shot' is the basic retrieval unit (last year there
was also the notion of 'clip'...) a common shot-set would be very useful in
our opinion - it also would simplify the evaluation of results.

We are of course willing to provide our segmentation data, but I'm not sure
wheter it would help much (the CWI segmenter is pretty braindead :)...

>3) Data distribution:  Right now we see shipping IDE disk drives as the
>   only practical way of sending 90GBs. We could create a pool of at 
>   least 4 and perhaps more drives, which we could FedEx to participants, 
>   who would copy the data and return the disk.
>   Alternatively, participants could send us an IDE drive,
>   which we could fill and return. Of course we have have some problems 
>   with drives due to shipping...
> 
>===> Anyone have a better idea?
>
>===> Assuming we use the IDE drives, how many current participants
>     think they would ship us a drive to fill?

We have downloaded the (proposed) dataset via FTP, but this was far from
a fun experience (using 6 connections at a time it took a few days...).

  (Note: we will followup later this week with a list of deviations found
  between file sizes of our downloaded files and the NIST-provided list;
  we are thinking of also reporting MD-5 codes for the files to enable
  checking of correctness)

Perhaps we could set up an exchange scheme, where groups 'share' their
data over the net on a more local basis, instead of downloading from the 
archive itself (assuming this does not violate restrictions from the archive)?
We will try to help out whenever needed.

Alex van Ballegooij & Arjen de Vries

====================================================================
CWI, room C0.11          Centre for Mathematics and Computer Science
Kruislaan 413                           Email: Arjen.de.Vries@cwi.nl
1098 SJ Amsterdam                       tel:       +31-(0)20-5924306
The Netherlands                         fax:       +31-(0)20-5924312  
===================== http://www.cwi.nl/~arjen/ ====================