Home Messages Index
[Date Prev][Date Next][Thread Prev][Thread Next]
Author IndexDate IndexThread Index

Re: MARS - Models of Appearance, Registration and Segmentation Feedback

  • To: Roy <roy@xxxxxxxxxxxxxxx>
  • Subject: Re: MARS - Models of Appearance, Registration and Segmentation Feedback
  • From: Roy Schestowitz <r@xxxxxxxxxxxxxxx>
  • Date: Thu, 23 Feb 2006 18:25:37 +0000
  • Delivery-date: Thu, 23 Feb 2006 18:32:58 +0000
  • Envelope-to: r@schestowitz.com
  • In-reply-to: <43FDD8B6.4020005@utc.fr>
  • Openpgp: id=74572E8E; url=http://schestowitz.com/PGP
  • Organization: MCC / Manchester University
  • References: <E1FBf67-0007qC-EP@banana.catalyst2.com> <43FBF203.1000505@schestowitz.com> <43FDD8B6.4020005@utc.fr>
  • Reply-to: s@xxxxxxxxxxxxxxx
  • User-agent: Mozilla Thunderbird 1.0.2 (X11/20050317)
Hash: SHA1

>>> Unfortunately, I did not manage to see any comments posted for
>>> proposed topics. Does it means that discussions are held offline or
>>> under ICQ or so?
>> I think it just means that no-one wants to comment. *frown* I guess
>> comment in sites or blogs are becoming a rarity, with so many such sites
>> around.
> There seems to be an appreciable number of visitors (>400 views per
> post), but people don't react that much. That's too bad considering the
> time you might have invested just in posting.

Yes, I know what you mean. It is nice to get feedback from readers, but
I am aware that what I write does get read, in due time. The more
popular block remains my earlier little-yet-growing project, which you
can find at http://schestowitz.com/Vision/

> I wondered if people weren't just reluctant to leak some information
> that could help competitors ;(
> I hope I'm really missing the point.

I used to leave comments in people's sites and blogs, but I rarely do so
anymore. I sometimes do, but not at the same capacity. It is hard to
track follow-ups and too many sites offer this facility, which was once
rather unique.

> Another explanation: AAM are an advanced concept, and seems quite hard
> to master. Maybe ASMs federate more people (like me!), maybe ASM
> problems might 'sound more familiar' to most people.

I must admit that I struggled with AAM's at first. Their applicability
and use is limited, due to implementational barriers and efficiency. It
was only recently that we found a valuable use for them, namely the
assessment on NRR, which requires no ground truth. Prior to that, they
were good for silly admiration and synthesis of faces (e.g. for video
games; see Genemation)

> Maybe it's just me, but even though I'm quite found of 3D PDMs, I felt
> like most topics were too AAM-oriented to really grab my understanding
> at first sight. But, hey, it's an AAM site...

The site initially revolved around AART (see relevant links), which is
Open Source and free to download, use and extend. It is a separate
project; actually it was a predecessor. Its aim was actually to attain
NRR using AAM's as an assessor. We are now reverse-engineering the
problem in some sense. We use AAM's to assess NRR.

> Maybe most readers are too much used to write serious publications that
> they are reluctant to write in a more casual style suiting blogs? Or
> maybe they would not like to say something stupid that could be read by
> a peer's audience...

The latter possibility if what I have always assumed.

>> I am not sure about /full/ annotation and identification of structures.
>> However, the papers from Davies (see cache at:
>> ) explain how to deal with shapes in 3-D and refine their model
>> representation.
>> Let me know how it goes...
> Yes, that's it ;)
> I'm already using the implementation of MDL designed by Allan Reinhold
> Kildeby, from IMM DTU. The contribution of R. Davies to this field is
> astounding!

I know one of the guys at IMM DTU (electronically). They do a lot of
work on AAM's and they also continue to extend much of the work that
originated here in Manchester. I cannot offer anybody else's code, but I
abide by this attitude of GPL'ing everything I write. If you need
something, give me a shout.

> Does automatic annotation makes sense for AAM applications? I imagine
> that facial 2D models do not need automatic landmarking, such as several
> 2D case studies. But for 3D ASM, that's just a question of survival...

To answer your first question: Yes. One of the things that we have been
working hard on throughout the past few years in fact) is automatic
landmarking. Subsequently, we can build /appearance/ models
automatically. We have done that with brain data and face data. As yet,
we have a caveat which is inter-individual face datasets.

> I must admit I'm really enthusiastic for the emergence of an informal
> forum where people could discuss freely about the problems (theoretical,
> implementation, etc) related to ASM-AAM. But, of course, I tend to
> project my own wishes & needs in the process!

I strongly believe that the facet of academic publications and
discussions is changing. In the past, people turned to journals when
they sought a solution to an arcane and very specific problem. Even
professors nowadays are pointing their students at Google *cough*. Mind
you, they don't even understand PageRank or equivalents, let alone pay
any attention to it. Technology helped many of today's kids teach their
parents 'this and that'. Often you will find that an IT-savvy lad such
as yourself will be required to help the professors, even when they are
shy. They fail to adapt to change, or simply refuse to do so, i.e. are
in a deep state of denial, yearning for the olden days.

What it boils down to is a scenario where publication on the Web is more
rapidly accessible. I estimate that thousands of people in my research
area visit my site every month, specifically for research-related
activities and interests. So what gives? Paper-form publications are
deem to reach a state of constant demise. They may be the 'currency' in
people's resumes, but they only collect dust on the shelves while
failing to get read very frequently, let alone be cited.

I am not in favour of this. I think it is rather sad, but I follow the
rules of the game. *smile*

Best wishes,


- --
Roy S. Schestowitz
http://Schestowitz.com  |    SuSE Linux     |     PGP-Key: 0x74572E8E
  6:00pm  up 6 days  6:19,  8 users,  load average: 1.08, 0.72, 0.60
      http://iuron.com - help build a non-profit search engine
Version: GnuPG v1.2.2-rc1-SuSE (GNU/Linux)


[Date Prev][Date Next][Thread Prev][Thread Next]
Author IndexDate IndexThread Index