[Gzz-commits] manuscripts/Paper paper.tex

[Tuomas J. Lukka]

CVSROOT:        /cvsroot/gzz
Module name:    manuscripts
Changes by:     Tuomas J. Lukka <address@hidden>        03/03/31 03:36:08

Modified files:
        Paper          : paper.tex 

Log message:
        twids

CVSWeb URLs:
http://savannah.gnu.org/cgi-bin/viewcvs/gzz/manuscripts/Paper/paper.tex.diff?tr1=1.76&tr2=1.77&r1=text&r2=text

Patches:
Index: manuscripts/Paper/paper.tex
diff -u manuscripts/Paper/paper.tex:1.76 manuscripts/Paper/paper.tex:1.77
--- manuscripts/Paper/paper.tex:1.76    Mon Mar 31 02:42:20 2003
+++ manuscripts/Paper/paper.tex Mon Mar 31 03:36:08 2003
@@ -45,12 +45,8 @@
 
 NO SEMANTIC CORRELATIONS!!!!!!!
 
-BLEACHING PIC + ZOOM EFFECT ON READABILITY
-
 MOTIVATING EXAMPLE NOT YET USER-TESTED
 
-IMAGE OF INVERTING THE VISUAL SYSTEM!!!
-
 IMAGE: RENDERING MODES, TEXTURE COORDINATES!
 
 SPARSE CODING: A TEXTURE CONTAINING BOTH YELLOW TRIANGLE AND
@@ -60,7 +56,7 @@
 TJL
 
 We present a perceptually designed hardware-accelerated
-algorithm for generating unique background textures for data.
+algorithm (on NV10, NV25) for generating unique background textures for data.
 To be recongizable,
 the texture should produce a random feature vector in the brain
 after visual feature extraction. 
@@ -98,14 +94,15 @@
 %In Focus+Context views, the textures can act as visual cues in the context
 %(information foraging).
 
-We discuss our freely available hardware-accelerated implementation 
-of unique backgrounds
-on the NV10 and NV25 architectures. 
+%We discuss our freely available hardware-accelerated implementation 
+%of unique backgrounds
+%on the NV10 and NV25 architectures. 
 
 We show the results of an initial experiment ... XXX
 
-Text readability is a major concern for using such backgrounds,
-and we discuss a method for enhancing readability by both providing 
+%Text readability is a major concern for using such backgrounds,
+%and 
+We discuss a method for enhancing text readability by both providing 
 fast, interactive zooming and 
 unnoticeably bleaching the background around text.
 
@@ -148,8 +145,6 @@
 % XXX icons
 \section{Introduction}
 
-TJL
-
 ``You are in a maze of twisty 
 little passages, all alike''\cite{zork}.
 It is easy to become disoriented and lost in an environment 
@@ -190,7 +185,8 @@
 of a displayed item at a glance, without explicitly reading the title.
 The user can even become aware of the identity just by seeing 
 any {\em fragment} of the item, instead of the title page. This property is 
vital
-for our example application discussed in Section~\ref{secbuoyoing}.
+for our example application discussed in Section~\ref{secbuoyoing} and is in 
fact
+one of the prime motivators for this work.
 
 
 % as a navigation aid in focus+context views.
@@ -216,9 +212,10 @@
 the BuoyOING focus+context
 user interface to a hypertext structure.
 Then, we formulate general principles for designing
-recognizable backgrounds and present a hardware-accelerated implementation.
+recognizable backgrounds, based on texture and image perception literature,
+and present a hardware-accelerated implementation.
 Following this, we discuss enhancing text readability on such backgrounds
-and practical experiences.
+and practical experiences and conclude.
 %Finally, we show an example application of unique backgrounds
 %for browsing linked PDF documents in a focus+context view.
 
@@ -711,9 +708,34 @@
 In this section, we discuss our hardware-accelerated implementation
 (libpaper)
 of unique backgrounds (papers).
-Our main platforms are NV10, i.e., OpenGL 1.3 + 
-\url{GL\_\hyp NV\_\hyp register\_\hyp combiners},
-and NV25, i.e., NV10 + \url{GL\_\hyp NV\_\hyp texture\_\hyp shader3}.
+
+
+\subsection{Parameter hierarchy}
+
+We have found that setting the parameters hierarchically
+produces the best results: the parameters for different passes
+should depend on hyperparameters randomly selected for entire paper.
+
+XXX REASONING: YELLOW SQUARE, RED TRIANGLE
+
+Our current parameter hierarchy is shown in 
+Fig.~\ref{figpipeline}. The individual parameters are
+explained in the subsections below.
+
+
+\begin{figure}[htbp]
+\ifpics
+\includegraphics[width=\linewidth]{pipe.1}
+\fi
+\caption{\label{figpipeline}
+The parameter hierarchy in our implementation.
+See text for explanations.
+}
+\end{figure}
+
+%Performance: zoomability, memory, speed
+
+\subsection{Rationale for a Fragment-based implementation}
 
 Our implementation needs to support complicated mappings
 between paper and screen coordinates, such as fisheye distortion.
@@ -721,6 +743,17 @@
 must be done on the fragment level after the texture accesses, i.e., we cannot 
use
 procedural geometry.
 
+Plain OpenGL 1.3 does not by itself provide enough flexibility in the fragment
+pipeline to allow the flexible generation of features.
+Because of this, and the availability of stable Linux drivers,
+our main platforms are NV10, i.e., OpenGL 1.3 + 
+\url{GL\_\hyp NV\_\hyp register\_\hyp combiners},
+and NV25, i.e., NV10 + \url{GL\_\hyp NV\_\hyp texture\_\hyp shader3}.
+We are working on an implementation based 
+on \url{GL\_\hyp ARB\_\hyp fragment\_\hyp program} once suitable hardware
+and Linux drivers emerge.
+
+
 % One important goal for the implementation is that it should be easily 
 % applicable in a variety of situations. 
 % For example, mapping between paper and screen can be complicated,
@@ -746,26 +779,10 @@
 requires 1 pass with only one texture unit.
 To achieve a satisfactory image quality in zooming
 with the latter approach,
-a relatively large texture has to be used.
+a relatively large texture has to be used; 512x512 is not really sufficient.
 
-We have found that setting the parameters hierarchically
-produces the best results: the parameters for different passes
-should depend on hyperparameters randomly selected for entire paper.
-Our current parameter hierarchy is shown in 
-Fig.~\ref{figpipeline}.
-
-
-\begin{figure}[htbp]
-\ifpics
-\includegraphics[width=\linewidth]{pipe.1}
-\fi
-\caption{\label{figpipeline}
-The parameter hierarchy in our implementation.
-See text for explanations.
-}
-\end{figure}
-
-%Performance: zoomability, memory, speed
+If using only the direct mode, it would be possible to use procedural geometry
+when generating the texture as well, but 
 
 
 \subsection{Colors}
@@ -954,6 +971,8 @@
 
 \subsection{Texture coordinates}
 
+XXX More
+
 % Texture coordinates define the mapping of the basis textures 
 % to paper coordinates.
 
@@ -976,6 +995,8 @@
 
 \subsection{Texture shading}
 
+XXX More
+
 On the NV25 architecture,
 the texture accesses can be customized further by the use of texture 
 shading: the texture coordinates used by a texture unit can be made
@@ -1248,6 +1269,9 @@
 a) Original,
 b) Zooming, c) Bleaching, d) The bleached version without drawing
 the black text, showing the blurred lightening.
+XXX [tjl] Jvk: MUCH less text, about a fourth of the amount; not two across 
but all vertically
+stacked. Several different gradations of b) and c). In c) also different 
sizes, one where
+the bleaching obviously shows the space between lines.
 }
 \end{figure}
 
@@ -1305,6 +1329,9 @@
 to its rate of accesses.
 The diagram shows 2000 documents weighted with Zipf's law with exponent 1.1.
 Here, the 15 most important documents account for approximately half of the 
accesses.
+Of course, it is impossible to know which documents will be important and 
+that will also change with time so all documents should be textured
+stably from the first viewing onwards.
 % 0.50469672124463749
 }
 \end{figure}
@@ -1361,6 +1388,8 @@
 The numbers give the 
 percentage of trials and the average reaction time in seconds for
 the five subjects.
+This shows that the textures are quite recognizable even after just {\em one}
+previous viewing.
 }
 \end{table}
 
@@ -1420,24 +1449,27 @@
 
 \section{Conclusions}
 
+We have presented a perceptually designed hardware-accelerated
+algorithm
+
 TJL
 
 We have introduced procedurally generated unique backgrounds
 as a way of visualizing the identity of data or documents.
-While the method is general and
-applicable whenever data has identity,
-it is at its best when the same data node can be reached through
-several ways, the nodes' overall appearance is similar,
-and fragments of nodes are seen. 
+
+The motivating example, the BuoyING user interface demonstrates
+that the method is at its most useful
+when the same document can be reached through
+several ways and fragments of documents are seen. 
 
 Of course, we cannot hope to match in quality a unique graphical
 appearance designed by a human designer; 
 magazines and web sites have long used skilfully designed 
 graphical elements to make themselves recognizable.
-Instead, the algorithm is able to generate an unlimited
+However, the algorithm is able to generate an unlimited
 amount of unique backgrounds cheaply, making it possible
-to give, e.g., all academic articles stored on a user's hard
-drive their own background.
+e.g. for all academic articles with similar typography
+to have their own background.
 
 %\subsection{Further work}
 
@@ -1447,9 +1479,10 @@
 as opposed to 2 and 2 on NV10, 
 using 2 passes as we currently do is too much; 
 it should be possible to obtain interesting textures with just one pass.
-We are also working on implementing
-these algorithms on OpenGL ARB extensions ..., due to their recent release
-of a Linux driver. 
+The more general ARB fragment program extension also ...
+%We are also working on implementing
+%these algorithms on OpenGL ARB extensions ..., due to their recent release
+%of a Linux driver. 
 
 % However, we see the proprietary extensions only
 % as an interim situation: the recently completed
@@ -1470,13 +1503,13 @@
 % to use repeating units for non-repeating backgrounds --- but use more than 
one,
 % which are not rationally related.
 
-In this article, we report initial 
+We report an initial recognition experiment
 
-Carrying out usability tests is necessary, both
+Carrying out usability more tests is necessary, both
 to measure how well textures can be remembered
 and to make the ad hoc distributions more experimentally based.
-It could also be worthwhile to experiment with other ways
-of visualizing identity, such as unique edge shapes.
+%It could also be worthwhile to experiment with other ways
+%of visualizing identity, such as unique edge shapes.
 
 XXX: the effect of zoomability on recognizability