Science and Technology news... :: View topic

Colm G. Connolly · Guest

Hi all,

Over the past two weeks or so I>ve been reviewing the work of David Marr and
that of the (inter)active vision researchers (the likes of Dana Ballard, J.
K. O>Regan and others) for my thesis. This lead me to wonder if it is
possible to reconcile the two different approaches to one theory.

My initial reaction is that they two the hypotheses come from opposite ends
of the representational spectrum: Marr>s theory is representation rich and
the active vision researchers tend to eschew representations and so hope of
finding suitable common ground is all but lost.

Is this a debate which has been largely bypassed or has it been directly
addressed?

Any comments or suggestions for further reading you have would be greatly
appreciated,

Thanks in advance,

--
_\\|//_
( O-O )
---------------------------o00--(_)--00o------------------------------
Colm G. Connolly |
Department of Computer Science |
University College Dublin (UCD) |
Belfield, Dublin 4 |
Éire / Republic of Ireland |

Glen M. Sizemore · Guest

It would be helpful if you would say a little bit about the two "camps"
especially the "active vision" one, since some people may know these under
different names. I can guess, though, and based on the guess say that, no,
there is no common ground. For example, for me, vision is a matter of visual
stimuli "turning on" responses. There is no intermediate, pre-behavioral
"act" of seeing, sensing, or otherwise having commerce with an alleged
representation. This is true even when a part of the response occurs later
when the original stimuli are not present (as in imagination).

"Colm G. Connolly" <colmconn@nowhere.nocountry> wrote in message
news:newscache$kq9slh$sq8$1@weblab.ucd.ie...
[quote]Hi all,

Over the past two weeks or so I>ve been reviewing the work of David Marr
and
that of the (inter)active vision researchers (the likes of Dana Ballard,
J.
K. O>Regan and others) for my thesis. This lead me to wonder if it is
possible to reconcile the two different approaches to one theory.

My initial reaction is that they two the hypotheses come from opposite
ends
of the representational spectrum: Marr>s theory is representation rich and
the active vision researchers tend to eschew representations and so hope
of
finding suitable common ground is all but lost.

Is this a debate which has been largely bypassed or has it been directly
addressed?

Any comments or suggestions for further reading you have would be greatly
appreciated,

Thanks in advance,

--
_\\|//_
( O-O )
---------------------------o00--(_)--00o------------------------------
Colm G. Connolly |
Department of Computer Science |
University College Dublin (UCD) |
Belfield, Dublin 4 |
Éire / Republic of Ireland |[/quote]

Joe Legris · Guest

Colm G. Connolly wrote:
[quote]Hi all,

Over the past two weeks or so I>ve been reviewing the work of David Marr and
that of the (inter)active vision researchers (the likes of Dana Ballard, J.
K. O>Regan and others) for my thesis. This lead me to wonder if it is
possible to reconcile the two different approaches to one theory.

My initial reaction is that they two the hypotheses come from opposite ends
of the representational spectrum: Marr>s theory is representation rich and
the active vision researchers tend to eschew representations and so hope of
finding suitable common ground is all but lost.

Is this a debate which has been largely bypassed or has it been directly
addressed?

Any comments or suggestions for further reading you have would be greatly
appreciated,

Thanks in advance,

[/quote]

Marocco, D. and Floreano, D. (2002) Active Vision and Feature Selection
in Evolutionary Behavioral Systems. In Hallam, J., Floreano, D. Hayes,
G. and Meyer, J. (Eds) From Animals to Animats 7. Cambridge, MA. MIT Press.

http://asl.epfl.ch/aslInternalWeb/ASL/publications/uploadedFiles/sab02-dm.pdf

--
Joe Legris

Mark Horn · Guest

25-SEP-2003

Dear Colm,

Since I>m not clear as to what precisely you>re aiming at when you speak
of "active vision," I>ll stress only the elemental significance of the
zero-crossings ("primal sketch") defined by the Marr-Hildreth model,
extended later by Koenderink and others. I think this is a reasonable
common ground from which one can reach any sensible theory.

There>s perhaps a good bit of mathematical physics to be squeezed out of
this form; for an interesting computational analysis search on Frederic
Guichard, and locate his doctoral dissertation, "Axiomatization of
Multiscale Analysis," University of Paris 1988; and another work he
wrote with his advisor Jean-Michel Morel, "Partial Differential
Equations and Image Iterative Filtering." Sorry, I have hardcopies
which bear no www. urls.

m.j. horn

David Longley · Guest

In message <newscache$kq9slh$sq8$1@weblab.ucd.ie>, Colm G. Connolly
<colmconn@nowhere.nocountry> writes
[quote]Hi all,

Over the past two weeks or so I>ve been reviewing the work of David Marr and
that of the (inter)active vision researchers (the likes of Dana Ballard, J.
K. O>Regan and others) for my thesis. This lead me to wonder if it is
possible to reconcile the two different approaches to one theory.

My initial reaction is that they two the hypotheses come from opposite ends
of the representational spectrum: Marr>s theory is representation rich and
the active vision researchers tend to eschew representations and so hope of
finding suitable common ground is all but lost.

Is this a debate which has been largely bypassed or has it been directly
addressed?

Any comments or suggestions for further reading you have would be greatly
appreciated,

Thanks in advance,

[/quote]
By "active" - do you mean "direct" as in the ecological approach
advocated by James Gibson and others who are anti-cognitivist in the
sense that they suggest you "ask what your head>s inside of rather than
what>s inside your head"?

--
David Longley

dan michaels · Guest

"Colm G. Connolly" <colmconn@nowhere.nocountry> wrote in message news:<newscache$kq9slh$sq8$1@weblab.ucd.ie>...
[quote]Hi all,

Over the past two weeks or so I>ve been reviewing the work of David Marr and
that of the (inter)active vision researchers (the likes of Dana Ballard, J.
K. O>Regan and others) for my thesis. This lead me to wonder if it is
possible to reconcile the two different approaches to one theory.

My initial reaction is that they two the hypotheses come from opposite ends
of the representational spectrum: Marr>s theory is representation rich and
the active vision researchers tend to eschew representations and so hope of
finding suitable common ground is all but lost.

Is this a debate which has been largely bypassed or has it been directly
addressed?

Any comments or suggestions for further reading you have would be greatly
appreciated,

Thanks in advance,

--
_\\|//_
( O-O )
---------------------------o00--(_)--00o------------------------------
Colm G. Connolly |
Department of Computer Science |
University College Dublin (UCD) |
Belfield, Dublin 4 |
Éire / Republic of Ireland |
[/quote]

Hi Colm, I>m cross-posting this over to c.a.p. Maybe can get John
Casey, who is actively engaged in computer vision research, and some
others, involved. I assume you>re interested in computer vision, but
you might also look at some of the work in wet neurophysiology of
vision - for background. Marr>s paper, especially, is a made-up
computational theory and probably not very close to emulating how the
brain solves the vision problem.

For vision neurophysiology, you might look at work by David Hubel +
Thorstein Wiesel, David van Essen, Charles Gray + Wolf Singer, Michael
Merzenich + John Kaas, and Semir Zeki. For vision psychology theories,
see Jerry Fodor and Zenon Pylyshyn. For vision system modeling, see
Gerald Edelman + Guilio Tononi, and also Francis Crick + Christof
Koch. The visual system in the brain involves massive feedback
connections between brain regions, and this aspect has hardly been
touched so far in wet research, hypothesized theories, or in
simulaton/modeling. Also, the human cortex devotes a large fraction
[maybe upwards to 20% - not sure???] of its area to vision -
indicative that the problem is not an easy one for the brain to solve.

dan michaels · Guest

"Colm G. Connolly" <colmconn@nowhere.nocountry> wrote in message news:<newscache$kq9slh$sq8$1@weblab.ucd.ie>...
[quote]Hi all,

Over the past two weeks or so I>ve been reviewing the work of David Marr and
that of the (inter)active vision researchers (the likes of Dana Ballard, J.
K. O>Regan and others) for my thesis. This lead me to wonder if it is
possible to reconcile the two different approaches to one theory.

My initial reaction is that they two the hypotheses come from opposite ends
of the representational spectrum: Marr>s theory is representation rich and
the active vision researchers tend to eschew representations and so hope of
finding suitable common ground is all but lost.

[/quote]

Colm, in addition to what I mentioned in the other post, follows is an
excerpt from Zenon Pylyshyn>s on-line book regarding the problem you
mentioned. manicmarvin mentioned this link on c.a.p, and I just
downloaded the 295 page [!!!] 4.5MB book this morning, so haven>t
looked thru it much as yet.

BTW, another Marvin - !Minsky! in this case - had some good words to
say about Pylyshyn>s ideas in his own on-line book - The Emotional
Machine:

http://web.media.mit.edu/~minsky/E5/eb5.html
[note - I tried checking this page just now, and it didn>t xfer
correctly - but the entire web>s been screwy lately]

Follows is from page 84 of Seeing and Visualizing: It’s Not What
You Think [2002], by Zenon Pylyshyn:

http://ruccs.rutgers.edu/faculty/pylyshyn/bookall.pdf

"... The active vision position is often presented as the claim that
vision should be characterized in terms of its potential for action
rather than its role in thought, and that when this is done we find
that representations become less important. It’s true that what
we see is determined to some extent by how we might potentially act
towards it and is also determined in part by the sensory-motor
contingencies we expect from such potential actions as eye movements
(see, for example, O>Regan & Noë, 2002). It has long been known that
the way the eye moves provides important information about the shape
of objects and contours (e.g., Miller & Festinger, 1977) and even
accounts for some illusions (see sections 3.4.1 and 6.5.2). It is also
true that if an object has a certain shape, the organism expects
certain sensory patterns to hold as the eye moves. If the eye moves
and these expectations go unfulfilled, a perceptual discordance
results. The discordance can lead to changes in perception, as it does
in the case of perceptual adaptation to distorting lenses. All this is
true and important. So it is true that vision depends not only on the
proximal visual stimulus (i.e., the retinal projection) but also on
patterns of perceptual-motor inputs and even on patterns of
expectations of these inputs. One can accept this without giving up on
the notion that what vision does is construct visual representations
which then serve in motor planning and in cognition the way beliefs
generally do, by entering into inferences that may or may not lead to
plans based on goals (i.e., old fashioned beliefs and desires).

Preparing for action is not the only purpose of vision. Vision is,
above all, a way to find out about the world, and there may be very
many reasons why an intelligent organism may wish to know about the
world, apart from wanting to act upon it. The organism may just be
curious (even rats and pigeons will respond for nothing more that the
information about when some non-reward-contingent light will come on,
and will do so with greater vigor than for food, e.g. Blanchard,
1975). In the case of humans, it is pretty clear that vision can serve
many functions including purely aesthetic ones. Indeed, so powerful is
the human drive for knowledge for its own sake that it led George
Miller to characterize humans as primarily “informavores”
(Miller, 1984)."

Colm G. Connolly · Guest

David Longley wrote:

[quote]By "active" - do you mean "direct" as in the ecological approach
advocated by James Gibson and others who are anti-cognitivist in the
sense that they suggest you "ask what your head>s inside of rather than
what>s inside your head"?
[/quote]
Well this was only one small part of what I had in mind. Sorry I should have
stated exactly what I meant by active vision. My "bible" on what
interactive vision is all about is "A critique of pure vision" by
Churchland, Ramachandran and Sejnowski in Large-scale neuronal theories of
the brain.

From my point of view I>m more interested in the neurobiological aspects of
vision.

If you take a look at Marr>s book, his theory is primely a highly modular
feed forward model where information is processed and representations are
created and only then can action take place. He would also seem to suggest
that vision is entirely passive: simply processing information before
passing it on to a separate system which controls action. This seems to be
fundamentally at odds with the neurobiology and some other experimental
work.

Another poster mentioned the massive feedback in the cortex, yet there is no
room for this in Marr>s theory, at least as I see it. Neither does the
brain seem to be as rigidly modular as Marr suggests, for example Peter
Lennie has a nice example of neurons involved in motor aspects of vision
being sensitive to colour as well.

Then there is our ability to initiate movement based on only scant
information, such as eye movements and head movements. In short we move to
see better. If we created a rich replica of the world, as Marr seems to
suggest, why would we need to do this. (I realise that this a somewhat of a
simplification but I hope you get the idea.)

Regards,

--
_\\|//_
( O-O )
---------------------------o00--(_)--00o------------------------------
Colm G. Connolly |
Department of Computer Science |
University College Dublin (UCD) |
Belfield, Dublin 4 |
Éire / Republic of Ireland |

dan michaels · Guest

"Colm G. Connolly" <colmconn@nowhere.nocountry> wrote in message news:<newscache$haszlh$xy6$1@weblab.ucd.ie>...

[quote]
Another poster mentioned the massive feedback in the cortex, yet there is no
room for this in Marr>s theory, at least as I see it. Neither does the
brain seem to be as rigidly modular as Marr suggests, for example Peter
Lennie has a nice example of neurons involved in motor aspects of vision
being sensitive to colour as well.
[/quote]

Actually, if you look at the work of Hubel+Wiesel, van Essen, Zeki,
and Merzenich+Kaas, among others, you>ll see that the visual parts of
the brain really are "highly" modular - with more than 30 different
cortical areas involved in different and quite specific vision-related
activities. However, the type of functionality discovered in these
areas probably isn>t very congruent with Marr>s theory, as it stands.

From what I have uncovered so far, Gerald Edelman is probably the main
guy who has directly tried to tackle what the massive feedback
pathways are up to - [although I>m sure there are other people+ideas
around]. This is totally open research territory. Marr>s theory is
very old now, and newer theories of biological vision need pay some
attention to the past 20 years of wet neuroscience research.

The Babynous Cult · Guest

dan michaels wrote:

[quote]"Colm G. Connolly" <colmconn@nowhere.nocountry> wrote in message news:<newscache$haszlh$xy6$1@weblab.ucd.ie>...

Another poster mentioned the massive feedback in the cortex, yet there is no
room for this in Marr>s theory, at least as I see it. Neither does the
brain seem to be as rigidly modular as Marr suggests, for example Peter
Lennie has a nice example of neurons involved in motor aspects of vision
being sensitive to colour as well.

Actually, if you look at the work of Hubel+Wiesel, van Essen, Zeki,
and Merzenich+Kaas, among others, you>ll see that the visual parts of
the brain really are "highly" modular - with more than 30 different
cortical areas involved in different and quite specific vision-related
[/quote]
ONLY THIRTY...???

do you know what these 'mapped' areas are looking at...???

[quote]
activities. However, the type of functionality discovered in these
areas probably isn>t very congruent with Marr>s theory, as it stands.

From what I have uncovered so far, Gerald Edelman is probably the main
guy who has directly tried to tackle what the massive feedback
pathways are up to - [although I>m sure there are other people+ideas
around]. This is totally open research territory. Marr>s theory is
very old now, and newer theories of biological vision need pay some
attention to the past 20 years of wet neuroscience research.[/quote]

dan michaels · Guest

The Babynous Cult <babynous@seanet.com> wrote in message news:<3F7C998D.9173E65@seanet.com>...
[quote]dan michaels wrote:

"Colm G. Connolly" <colmconn@nowhere.nocountry> wrote in message news:<newscache$haszlh$xy6$1@weblab.ucd.ie>...

Another poster mentioned the massive feedback in the cortex, yet there is no
room for this in Marr>s theory, at least as I see it. Neither does the
brain seem to be as rigidly modular as Marr suggests, for example Peter
Lennie has a nice example of neurons involved in motor aspects of vision
being sensitive to colour as well.

Actually, if you look at the work of Hubel+Wiesel, van Essen, Zeki,
and Merzenich+Kaas, among others, you>ll see that the visual parts of
the brain really are "highly" modular - with more than 30 different
cortical areas involved in different and quite specific vision-related

ONLY THIRTY...???

do you know what these 'mapped' areas are looking at...???

[/quote]
You can start here:

http://webvision.med.utah.edu/VisualCortex.html#pathways
http://white.stanford.edu/~heeger/psych202/lecture-notes/
visual-cortex/visual-cortex.html


www.GetXFactor.com Leading Technology, Science, Agriculture News and information	Part of the Identityscape.com network... getxfactor.com jmoodmusic.com smartbusinesschoices.com mintdepot.com lowfaresalways.com evangelicalview.com shoppingpodder.com soproudlywehail.com webnews.ws currenthumor.com

Science and Technology news... Forum Index -> Cognitive Science Forum
Page 1 of 1