[Gzz-commits] gzz/Documentation/Manuscripts/Paper paper.tex S...

[Tuomas J. Lukka]

CVSROOT:        /cvsroot/gzz
Module name:    gzz
Changes by:     Tuomas J. Lukka <address@hidden>        02/11/20 02:09:21

Modified files:
        Documentation/Manuscripts/Paper: paper.tex 
Added files:
        Documentation/Manuscripts/Paper: SCRATCH 

Log message:
        Add scratch

CVSWeb URLs:
http://savannah.gnu.org/cgi-bin/viewcvs/gzz/gzz/Documentation/Manuscripts/Paper/SCRATCH?rev=1.1
http://savannah.gnu.org/cgi-bin/viewcvs/gzz/gzz/Documentation/Manuscripts/Paper/paper.tex.diff?tr1=1.42&tr2=1.43&r1=text&r2=text

Patches:
Index: gzz/Documentation/Manuscripts/Paper/paper.tex
diff -u gzz/Documentation/Manuscripts/Paper/paper.tex:1.42 
gzz/Documentation/Manuscripts/Paper/paper.tex:1.43
--- gzz/Documentation/Manuscripts/Paper/paper.tex:1.42  Tue Nov 19 15:55:15 2002
+++ gzz/Documentation/Manuscripts/Paper/paper.tex       Wed Nov 20 02:09:21 2002
@@ -72,27 +72,39 @@
 
 \subsection{Focus+Context views}
 
-Focus+Context (fisheye) views\cite{fc-fisheye} are one paradigm for viewing 
large, 
+Focus+Context, or, fisheye views\cite{fc-fisheye} are 
+a paradigm for viewing large, 
 structured information sets 
 which has gained popularity in the research community\cite{XXX} 
 but less so in production systems.
 
-The basic principle in Focus+Context views is that of showing the current area 
of 
-interest (focus) magnified, and the structurally connected elements with less 
magnification.
+The current area of 
+interest (focus) is shown magnified, 
+and the structurally connected but further-away 
+elements with are shown peripherally, with less magnification.
+
+We distinguish two basic modes of application:
+\begin{enumerate}
+\item exploring data that the user
+has not seen before.  The research questions asked are, for example, 
+how fast the user can find a particular node in an unknown
+information hierarchy\cite{pirolli01informationscent,XXX}.
+\item everyday use of the user's own, familiar data.
+This use is seen for example in \cite{mackinlay91perspectivewall}.
+\end{enumerate}
 
+In this article, we focus on 
 
-XXX f+c refs, write
-
-Focus+context\cite{fc-fisheye,fc-taxonomy,fc-hyperbolic}
-
-Structural focus+context: not 2D map or so; ZUI are a borderline case.
-
-The method presented in this article is at its most useful in a particular 
kind of focus+context view:
-
-In this article, we 
-
-Visual information foraging in a focus + context visualization: information 
scent\cite{pirolli01informationscent}
-
+% XXX f+c refs, write
+% 
+% Focus+context\cite{fc-fisheye,fc-taxonomy,fc-hyperbolic}
+% 
+% Structural focus+context: not 2D map or so; ZUI are a borderline case.
+% 
+% The method presented in this article is at its most useful in a particular 
kind of focus+context view:
+% 
+% In this article, we 
+% 
 
 \subsection{Data Identity}
 
@@ -520,248 +532,8 @@
 \item Papertest, showing many images of papers, zooming, ...
 \end{itemize}
 
-\if0
-
-
-
-
-\section{old}
-Texture is an important visual attribute: it is used in the human
-visual system for e.g.~image segmentation and motion perception.
-So far, the most important uses of texturing in HCI have been 
-in 3D realism\cite{QuakeIII}
-and shape perception\cite{XXX}.
-
-% The human eye detects motion and zooming by XXX 
-% Texture helps to remove the local ambiguity
-
-% Often small, repetitive texture maps are used, due to 
-
-% Identity...
-
-% Desktop computers now have enough graphics performance to 
-% consider...
-
-In computer games, similar objects often appear not only similar but exactly
-the same: with the same texture and no noise, the objects 
-give an unrealistic appearance due to the sameness.
-
-Similar-but-different texturing has long been a hallmark of high-quality motion
-picture computer graphics.
-There, the need for realistic high-resolution textures 
-induced the invention of {\em Perlin noise},
-a repeatable band-limited random function. 
-This function looks similar everywhere in
-its domain but the actual values depend on the location. 
-The same quality is
-carried over to other functions whose values at each point are calculated
-by using noise in same way. Textures generated by such computations
-are called 
-{\em procedural 
textures}\cite{texturing-and-modeling,renderman-spec,advanced-renderman}.
-
-The above ideas about similarity and identity
-prompted us to ask whether that recognizability of ``sameness''
-could be put to use
-in user interfaces, by making an effort to texture similar but different
-things {\em differently}, coupling the texture of the object to its
-{\em identity} rather than its class.
-For example, if all documents were to have an individual, procedurally 
generated
-background, recognizing 
-a piece of the document shown as a 
gloss\cite{zellweger99:_fluid_links_infor_increm_link_trans}
-could be considerably faster, even subconscious.
-
-% This shares the goal of {\em brand identity} in advertising: similar products
-% are differentiated by graphical elements specific to the different brands.
-
-In the following sections, we first
-    discuss the context of this work: the Xanadu model, 
-    and then the tools used to create unique textures,
-    and our implementation on the NV20 architecture.
-Finally, we discuss 
-    our user experiments
-and 
-    why these textures should be easy to recognize.
-
-\section{Identity; the Xanadu model}
-
-Our interest in identity and its visualization stems from our exposure
-to Nelson's Xanadu model\cite{ted-xanalogical-structure-needed} in connection
-with our Gzz\cite{gzz} project.
-
-In the Xanadu hypertext model, the {\em identity} of a piece of media (text, 
image,
-video, sound) is raised to first-class citizenship. 
-A letter in a text is not just a character (byte value 0-255 or 0-65535) but 
rather
-the particular instance of the character that was typed on a certain keyboard 
at a certain
-point in time. This identity is permanent and remains across cut\&paste.
-The identity is not tied to the particular copy but the entry of the character 
to the system.
-
-Retaining the identity in this way has deep implications: for example, cutting 
and
-pasting small pieces of many documents to one document (for example, points 
relevant
-to certain talk) actually \emph{creates an implicit hyperstructure} which
-can later be traversed. The Xanadu model includes the ability to efficiently
-view all other documents that share some of the media with a given document.
-
-The context of this article is the use of focus+context views for showing
-connections: e.g.~if the current document is linked to a PDF file, the relevant
-piece of the PDF file could be shown shrunken in the margin, and moving to it
-would expand the whole file. In this situation, if the quoted piece is 
relatively
-small, using a unique texture for each PDF file might aid navigation: a 
familiar
-article would be recognized by the texture even from the small piece, without 
thinking.
-
-
-\section{How to draw paper}
-
-\subsection{Goals}
-
-\begin{description}
-\item[Distinguishability]
-    The papers generated with different seed values should be easily
-    distinguishable based on colors and texture. 
-\item[Speed]
-    Rendering the paper should be fast on modern graphics hardware.
-    We consider three alternative profiles: G400 on XFree86 DRI, and NV10
-    and NV20 with NVIDIA's proprietary drivers.
-\item[Memory]
-    Texture memory should be used parsimoniously.
-\item[Zoomability] 
-    Performance should not be affected by zoom factor.
-\item[Readability]
-    It should be possible to read text on top of the paper; the colors
-    need to be chosen with this in mind.
-\end{description}
-
-\subsection{Useful primitives}
-
-In this section 
-
-\subsubsection{Texturing}
-
-If the paper requires geometry to be actively drawn (e.g.~by
-having a grid which is drawn onto it as narrow polygons), then
-performance may suffer when the user zooms out far enough. 
-
-The system described in this article works by rendering a single large
-polygon and is therefore not affected.
-
-\subsubsection{Pseudo-random numbers}
 
-Instead of storing a database of millions of papers, we generate
-the papers procedurally\cite{texturing-and-modeling} by using
-repeatable sequences of pseudo-random numbers generated from a seed.
 
-\subsubsection{Noise, turbulence}
-
-\emph{Perlin noise} \cite{perlin-noise-intro}.
-is a repeatable band-limited random function that is fast to calculate.
-However, it is still outside the scope of what can be done on 
-current texture shaders; the next generation of hardware will likely change 
this.
-However, we do make use of pre-rendered noise in textures. This considerably
-limits the possibilities but is still useful.
-
-Rewrite: Fourier synthesis; different tradeoffs when pregenerating
-textures!!!
-
-\subsubsection{3D texture}
-
-The shape in a 2D texture is relatively fixed. A texel always has the same 
neighbouring
-texels, regardless of distortion. With 3D textures, three instead of two 
texture
-coordinates are used to access the texture. This allows the same texels to 
occur in
-different contexts, allowing different shapes to come forth from the same 
texture.
-
-XXX: rational vs irrational coordinates, shifting.
-
-\subsubsection{Related texture coordinate generators}
-
-If all texgens are chosen independently, the final texture will appear to
-consist of different layers and not repeat. In order to increase the different
-shapes in the textures, we can bind the $xy$ repeat units of two textures
-to each other even though there is no relationship between the textures. This
-creates a single repeating texture that is a combination of the two textures.
-
-This creates shapes that repeat exactly and are therefore easy to learn for 
humans.
-
-\subsubsection{Texture shading}
-
-Only available in the latest generations of consumer-level graphics
-cards, 
-
-Helps by making textures less recognizable
-
-Not all texture shading operations are suited equally well for
-the task at hand; most operations are aimed at lighting etc.
-
-Having a small texture act as a color map for a large texture ...
-
-Creates new shapes.
-
-Our experiments have been performed on the NV20 and NV25, 
-
-\subsubsection{Texture environments / Register combiners}
-
-The texture shaders (or without shaders, the texture accesses)
-provide a set of RGBA values to the next stage of the pipeline
-where these values are combined to form the fragment that gets 
-blended to the framebuffer.
-
-A flexible framework such as the register combiners of the NV10 and NV20
-provides an opportunity to use the RGBA values coming from the 
-texture in unconventional ways. Before the advent of texture
-shaders, register combiners were used for 
-calculating a dot product of the texel with a light vector at this stage
-in order to obtain bump mapping.
-
-For our purposes, the it is useful to decouple the ``colors'' in the textures
-and the 
-
-
-\subsubsection{CIELAB color space}
-
-A perceptually uniform color space is vital for choosing colors compatible to 
at
-least some degree.
-
-\subsection{Putting it together}
-
-\section{Implementation}
-
-
-\subsection{Profiles}
-
-Different OpenGL cards have different capabilities: numbers of textures, 
texture
-shading, blend operations etc.
-
-
-It would be good if the paper for the same $k$ would look exactly the same
-everywhere. However,
-since different hardware has different capabilities tradeoffs must be made
-in order to avoid using only the lowest-common-denominator capabilities.
-While it is easy to use the same overall color scheme on different hardware,
-the use of texture shading is possible only on high-end hardware.
-
-As a reasonable tradeoff, we introduce {\em profiles} which are
-sets of OpenGL features and extensions used; on the same profile, the
-same $k$ will always give the same paper (possibly at a different
-resolution). In our initial implementation there are three profiles:
-two textures, no shading, no special texture environments;
-N textures, product texture environment, XXX;
-4 textures, texture shading. These correspond to G400,  GeForce2 and
-GeForce3.
-
-
-
-\section{Conclusions}
-
-Jvk: identity as random vector, it being a texture is less important
-
-...
-Unfortunately, we were not able to apply the texture shading
-ATI's latest cards because ATI has not provided XFree86 drivers for those
-features.
-
-Especially zooming ...
-
-Counter: icons, keywords, readability
-
-\fi
 
 \bibliographystyle{plain}
 \bibliography{gzigzag}