|
Reviving/summarising Mail
|
stephane.marc-@cui.unige.ch
|
Oct 18, 2001 05:17 PDT
|
Dear Benchathletes,
At last! It has been some time of low traffic on this list. I hope you
all had a relaxing summer and a not-too-busy start of Academic year (for
academics) and post-summer period (for companies).
This mail is for asking to all of you at what stage you think we are
and we you personnaly are within the task you were working on. I remind
you that for now our priuority is to get a basic benchmarking setup
working for demonstration and advertisement at SPIE in January. This
means that by then we should have accomplished some steps in the
following tasks:
* Define a test image collection: Hui was working on gathering a free
collection, could youplease send a mail summarizing the state of the
collection? I know that Corinne and Peter are working on defining a
subset for peforming an initial annotation effort (please comment).
* Annotating the collection: The idea is to associate annotation
(keywords) to the above subset. Corinne has been working on a simplified
vocabulary. Thomas has successfully worked on the annotation tool in
PHP. The vocabulary is integrated and the tool is ready to be
distributed (we will include it in the SourceForge directory). If we
were to provide you a set of image with a set of keywords, how do you
think you would prefer going about associating both: with a tool such as
that at (http://viper.unige.ch/annotate - user test / passwd test) or
would you do it by hand or... ??? (please respond)
* Concretising the benchmark architecture: The idea is define the
harness to include any system within our framework. Wolfgang and Neil
have been working on that. What is the status right now (Wolfgang, a
short description and tutorial on the harness?)?
* Derive Relevance judgements from annotations. Any ideas,
recommendations (based on the type of annotations we wish to obtain)?
* Define performance measure. For that, we will base our effort on what
has been written in the literature (BIRDS-I and Viper's benchmarks).
* Other issues?
It seems relatively critical that something from the Benchathlon is
demonstrated at SPIE. We therefore envisage a fallback solution in the
form of the Viper benchmarking architecture (Henning, a reminder about
this architecture?). This benchmark shows the feasibility and
demonstrates the target architecture. It will constitute the block-set
from which we are aiming are replacing elements...
Please do respond and feel free to distribute this mail.
Best regards,
Stephane
--
Stephane Marchand-Maillet, PhD
Computer Vision Group, Universite de Geneve
24 Rue du General-Dufour - 1211 Geneva 4 - Switzerland
Tel: +41 (0)22 705 7631
Fax: +41 (0)22 705 7780
Email: Stephane.Marc-@cui.unige.ch
URL: http://viper.unige.ch/~marchand
http://viper.unige.ch/
http://www.mrml.net/
|
|
 |
|
