Dear All,
I'm new to birding-aus and have some comments questions about this.
In my non-birding life I work on computer vision, having just moved to
Adelaide from Oxford two months ago. I know Pietro Perona who is the
computer vision lead and have had some discussions with him about the
project. I have also birded once or twice at conferences with one of
the postdocs associated with another group who collaborate, Ryan
Farrell. If you are interested in the state of the art then as well as
checking out Pietro's site, also look at Ryan's paper from last year's
International Conference on Computer Vision at this web page
http://www1.icsi.berkeley.edu/~farrell/
The state-of-art is some way from being able to identify birds in
general images, or even from nicely posed photographs. Most human
biometric stuff that actually works relies on quite controlled
conditions for acquisition of the imagery (e.g. passport photo
recognition), including the pose and lighting. These are two key
factors that make identifying a bird in a photo or video really tough.
Furthermore it is an open research question to what extent one needs to
program expert knowledge into a machine recognition system for this kind
of subordinate categorization task, versus being able to learn it from
training data. The latter has met with some stunning successes in
recent times, and in terms of general object class recognition computers
can do reasonably well on 10s or sometimes 100s of distinct classes, but
there is no clear answer as to how far this can be pushed when trying to
distinguish very closely related subordinate classes.
This area of so-called "Fine-grained Visual Categorisation" is one that
interests me and I would love to think there is a will within Australia
to do something similar to the Cornell/Caltech, so it would be great if
there is any advice from this list that people could offer. In partic,
(i) is there any equivalent to Cornell (or the Edward Grey Institute in
Oxford) in Australia that would have an interst in engaging in such a
project; (ii) what databases of photographs exist and how can they be
used/accessed/etc; (iii) would there be any will from the birding
community at large to participate in markup of a database with a view to
building an identification tool for Australian and/or Australasian birds?
Cheers, Ian
On 15/11/2012 10:04 AM, Carl Clifford wrote:
Dear B-A,
Cornell Lab of Ornithology is developing a project called Merlin. This aim of
this project is to develop software that enables computers to identify birds
from images. See http://dev.nabirds.org/Web/Tools/ImageUpload/pages/about.php
Given the advances in human biometrics, I imagine that computer ID of birds is
quite doable. Wonder how long it will take for the technology to appear in
handheld devices? After all, how many of us say, 10-15 years ago, would have
thought that an all singing, all dancing field guide in a smartphone type
device would be possible?
An interesting space to watch.
Cheers,
Carl Clifford
===============================
To unsubscribe from this mailing list,
send the message:
unsubscribe
(in the body of the message, with no Subject line)
to:
http://birding-aus.org
===============================
--
Prof. Ian Reid
School of Computer Science
University of Adelaide
Adelaide, 5005
ph: +61 (08) 83132135
www: http://cs.adelaide.edu.au
===============================
To unsubscribe from this mailing list,
send the message:
unsubscribe
(in the body of the message, with no Subject line)
to:
http://birding-aus.org
===============================
|
