[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Gzz] address@hidden: GI 2003 notification - #182] -- sigh
From: |
Tuomas Lukka |
Subject: |
[Gzz] address@hidden: GI 2003 notification - #182] -- sigh |
Date: |
Mon, 24 Feb 2003 18:33:43 +0200 |
User-agent: |
Mutt/1.4i |
Blach... seems I and jvk screwed up in thinking that we could publish
at this conference without the relevant experiments. Well, live and learn...
I.e. *never* attempt to publish if all the parts you feel would be needed
aren't there.
Tuomas
----- Forwarded message from address@hidden -----
Date: Mon, 24 Feb 2003 10:43:29 -0500
From: address@hidden
Reply-To: address@hidden
To: address@hidden
Subject: GI 2003 notification - #182
CC: address@hidden
X-Virus-Scanned: by AMaViS-perl11-milter (http://amavis.org/)
Dear Tuomas Lukka -
We regrent to inform you that your paper
182 - Representing Identity by Unique Background Textures
has not been accepted to GI 2003.
Out of the 96 submissions 32 were accepted.
The reviews are included below.
Sincerely,
Torsten Möller and Colin Ware
Graphics Interface co-chairs
address@hidden
address@hidden
---------------- GI 2003 paper 182, review 1 ----------------
Title: Representing Identity by Unique Background Textures
Overall rating 2 (scale is 1..5; 5 is best)
Reviewer expertise 2 (scale is 1..3; 2 = "Knowledgable")
1. Rating
2 (Perhaps reject)
2. Expertise
2 (Knowledgable)
3. The Meta-Review
Meta-Review:
Reviewers said essentially 'cute, but does it work?'. Strength:
interesting and novel UI method, nice algorithms for hardware-based
implementation. Concerns: no evaluation so it's hard to believe it would
actually be helpful, method is 'ad-hoc' rather than principled, hard to
scale given interaction between spatial frequency and color perception.
The textures used seem visually intrusive. It would be difficult to
imaging that such strong colors would be desirable as a background to
text.
The committee decided to reject this paper because of concerns about
technical content, the insufficient justification/motivation, and the
lack of evaluation. On the content side: despite all the discussion of
how the authors strove to maintain legibility, the high-order bit for
most of the committee was "I would have a *really* hard time reading
that document!". There has been a lot of work on texture perception that
was completely ignored - most relevant is the issue of texture
discriminability. The fact that color perception changes as the texture
size changes (with viewpoint manipulation) was also completely ignored.
The paper would benefit from more motivation of when this technique would
be useful - presumably, only when parts of documents are actually viewed
side by side, because it's highly unlikely that the background texture
could be compared to a remembered one from a document seen earlier. So
this approach would not be useful for the obvious case of the document
space of the World Wide Web. Evaluation would also be very useful.
One alternate suggestion on approach if this paper is resubmitted would
be to deemphasize the UI focus as being too speculative, and change the
focus (and title) to the texture algorithm as a graphics contribution.
personal review:
Strong and interesting work - original and clever idea, the results
were compelling, the description was clear. Although there are obvious
concerns about legibility, the authors do discuss at length how their
design choices minimize this problem.
low-level:
Figure 6 - replace XXX with [23]
section 4.2 sentence 1 - comma between 'textures' and 'Even' should be
period.
section 2.1 - Interrante's work would also be a good thing to mention
at the end of the texturing related work section (where you currently
cite Schweitzer and Ware). For example: Victoria Interrante
"Illustrating Surface Shape in Volume Data via Principal
Direction-Driven 3D Line Integral Convolution", SIGGRAPH 97
---------------- GI 2003 paper 182, review 2 ----------------
Title: Representing Identity by Unique Background Textures
Overall rating 3 (scale is 1..5; 5 is best)
Reviewer expertise 2 (scale is 1..3; 2 = "Knowledgable")
1. Rating
3 (Neutral)
2. Expertise
2 (Knowledgable)
3. The Review
1) This paper describes a low-end hardware-accelerated implementation of
a graphical system for identifying documents with unique background
textures. The textures are derived from each document's hash code along
with some heuristically selected palette of colors for each document.
2) It is interesting to note that a perceptual model was considered for
the creation of unique textures. The algorithm devised is fast and
'cheap' in producing the unique textures.
3) The paper is very appropriate for GI but could be slightly improved if
it were to get published. It lacks in the following areas:
- the description of the heuristics used for the color selection of each
document could be expanded. The paper does not suggest what set of
heuristics are applied
- requiring that the display gamma be properly adjusted seems to be a
strong handicap in the implementation. The authors do not suggest what
effects would be created if the display gamma is not adequately
corrected.
- the explanation in figure 6 does not lead the reader to the appropriate
interpretation. What is suggested in the caption is not clearly apparent.
This figure seems central in showing the practical value of the results
of the algorithm.
- it would be interesting to know how 'unique' are the textures when
somewhat similar documents are are used.
- the research could also benefit from a user evaluation.
4) The content and structure are well presented. However, certain
sections could be reduced to a couple of lines or to one paragraph (and
replaced with content as described above). For example, the background
material on texturing can be reduced to one/two paragraph(s), and,
similarly the section on focus+context does not need as extensive an
explanation since it is only used for displaying one sample output. There
are several typos and in some cases references are missing. For example,
reference to Nelson in figure 6 is missing ([23]). Also, the style used
for references is not consistent (see refs 9-10, and then 10-11).
---------------- GI 2003 paper 182, review 3 ----------------
Title: Representing Identity by Unique Background Textures
Overall rating 4 (scale is 1..5; 5 is best)
Reviewer expertise 2 (scale is 1..3; 2 = "Knowledgable")
1. Rating
4 (Perhaps accept)
2. Expertise
2 (Knowledgable)
3. The Review
What is the main point of this paper?
This paper proposes a system to automatically generate background
textures for items so as to assist in their identification as particular
elements (or types) when used in a focus+context system. A system
illustrating this is implemented on NV10 and NV25 architectures, and
shown to run reasonably quickly.
What are the significant and novel contributions of this paper to the
field of GI?
This system is an interesting one, in that it looks at whether textures
can form the basis of quick identification in a focus+context system. I
have a few minor concerns - e.g. Im not sure that adequate 3D shape of
the surface can be recovered from the texture if luminance differences
are small (which would limit the maximum possible dimensionality
somewhat). But I think the general idea is an interesting one and
deserves to be explored.
In your opinion is this paper appropriate for GI 2003?
Yes
Is this paper adequately written (content, structure, English)?
Yes
---------------- GI 2003 paper 182, review 4 ----------------
Title: Representing Identity by Unique Background Textures
Overall rating 1 (scale is 1..5; 5 is best)
Reviewer expertise 2 (scale is 1..3; 2 = "Knowledgable")
1. Rating
1 (Definitely reject)
2. Expertise
2 (Knowledgable)
3. The Review
The paper proposes a way to create file-specific textures to identify
file content. The goal is to help make it easy to identify identical
documents when they appear in different settings. Their main example was
multiple copies of an article stored on a computer under different
names--can you tell by looking at the background that they are the same?
The algorithm can be implemented in hardware.
The paper has very grand goals, but presents only an ad hoc result. Right
at the point where they needed to prove their technique has any value,
they launched into a detailed description of how to implement it. The
questions this approach asks are many:
--How well does it work? Any trials, any studies, any anecdotes?
--Do similar files look similar? Was this a goal?
--How does this technique scale? Can you really distinguish and remember
all these colored textures among, say, 100 different files? How much
better is it really than color coding alone?
--What happens when you scale the textures, for example, as they recede
along the document wall in Figure 6? Color perception is strongly
affected by spatial frequency, and shapes will lose their detail as the
view shrinks. Do you really see the small and the large versions of the
texture as "the same." Obviously you won't in the limit.
To my eye, the textures are overly bright, bold and colorful, both on my
screen and on my inkjet printer. This is consistent with their statement
that "the saturations are chosen from distribution emphasizing saturated
colors" (section 4.1). On the other hand, they claim to be using high L*
values, much lighter than I see in the PDF file viewed on my display.
I'm running a PC system with a 2.2 gamma. The text is perfectly visible,
but the background colors and textures far too prominent for comfortable
reading, IMHO. They need to say what gamma value they designed for if
they want a careful visual evaluation.
I find Figure 6 very confusing. What are the little call-outs pointing
into the document wall? Are they part of the UI, or just part of the
illustration? Does the user normally have both the top and bottom view
open simultaneously? Unfortunately, I was unable to view their
animation, even with the latest version of the Windows Media player for
Windows 2000. Perhaps the animation would have helped clarify the
application, but the paper should stand without it.
I recommend rejecting this paper as a UI paper because it has no proven
UI value of any subtlety. The hardware implementation of a texturing
algorithm with their specific parameters and constraints may be
interesting for other purposes (that's not my field), but it isn't the
stated contribution.
---------------- GI 2003 paper 182, review 5 ----------------
Title: Representing Identity by Unique Background Textures
Overall rating 2 (scale is 1..5; 5 is best)
Reviewer expertise 2 (scale is 1..3; 2 = "Knowledgable")
1. Rating
2 (Perhaps reject)
2. Expertise
2 (Knowledgable)
3. The Review
The authors developed some sophisticated techniques for procedurally
generating textures for the backgrounds of objects so that users could
recognize them.
The application is a focus+context document viewer where samples from
document B can be popped up while viewing document A. The texturing
technique is applied to the paper of each document. The user would
presumably recognize the backgrounds of documents A and B.
The main difficulty I have with the paper is that the technique is in a
vacuum. There seemed to be little evaluation of the technique, certainly
no experiment or formal user test was reported.
I don't really know important a problem the authors have solved.
It seems to rest on the assumption that document readers will recognize
the texture but not the text of a document.
The authors speak at length about the texture intensity issues that this
system provokes. It would be interesting to see a more structured
analysis of these issues. As it stands, the authors' approach is mainly
to tweak the process until it is acceptable, but there is a lot more
going on there that might be amenable to a signal analysis approach that
the authors should investigate more formally.
----- End forwarded message -----
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [Gzz] address@hidden: GI 2003 notification - #182] -- sigh,
Tuomas Lukka <=