There is no one best / correct way to build these systems -
choices are made based on use cases, available space, and
available $.
Some of the largest displays currently in use
TACC
Stallion
2D 328 MPixel wall (40,960 x 8000) - made from 80 Dell 3007WFP
Widescreen LCDs customized to reduce borders
CPU - 40 X 6-core Intel Sandy Bridge 2667W @ 2.9 GHz
RAM - 64GB per node
Disk - 1TB per node
GPU - 1 NVIDIA GTX 680 per node
Lasso - 2D Multitouch 12.4 MPixel (5760x2160) - made from 6 (3 x
2) 46" monitors
Single computer
Reality Deck
2D 1,500 MPixel - made from 416 Samsung S27A850D 2560 by 1440 LCD
displays covers four walls of a 33' x 19' x 10' room
CPU - 18 X dual 6-core Intel Xeon
RAM - 48 GB Per Node
Disk -1 TB per node
GPU - 4 AMD Firepro V9800 GPUs (each node can drive up to 24
displays)
building a 24.4 soundsystem
Its
not just about having the biggest display - big displays are
great, they push the boundaries and get us closer to a future
where any surface could be a display if that's what we want, but
they take a lot of space and cost a lot of money to build and
maintain. They are often built in visualization centers
surrounded by visualization people, but not necessarily
surrounded by domain scientists or other potential users.
To be
really useful these kinds of displays need to be near their
users so those users have regular easy access to them. Users do
not want to walk across campus to a visualization center. They
don't want to walk to another building. They often don't even
want to walk to another floor of the same building. This makes
smaller but still high-resolution large-format displays in
conference rooms or labs more appealing.
here is a web page showing many of the various smaller displays
running the same software the we use during class - sage
http://www.sagecommons.org/community/sage-sites/
There are portable versions of these displays that can be moved
around - Optiportables
http://www.calit2.net/newsroom/article.php?id=1808
Projection-based systems
advantages
can use edge blending to create a pretty
decent single big image so there are no borders
can make very large images
can project onto irregularly shaped
surfaces
can pretty easily make a front projection
floor
disadvantages
hot and noisy - can be dealt with through
vents and shielding but its additional cost
room lighting needs to be dim or dark for
best results - and dark adaption time can be important
color and contrast not as good as LCDs
without buying very expensive projectors
projectors need to be colour calibrated and
aligned semi-regularly
bulbs need to be replaced every few
thousand hours - this is improving with newer projectors
projectors do not turn on as quickly as
LCDs and are usually left off when not in use to conserve the
bulbs
bright high end projectors physically take
up a lot of space
throw distance required between the
projector and the screen
Rear
projection systems mount the projectors behind a black screen.
Front projection systems mount the projectors on the ceiling in
front of the screen. The advantage of rear projection is that
the users can get very close to the screen without casting
shadows and the noise and heat of the projectors can be hidden.
The disadvantage is that the projectors take up a lot of wasted
space to have enough throw distance behind the screen, even if
the optics are folded with mirrors.
Front projection systems have the advantage of being able to
project onto existing surfaces, although projecting onto
dedicated front projection screens gives better brightness and
contrast, and is required for passive stereo projection. Very
large screens are expensive and also have seems.
Front projection screens are best in theatre situations where
the users will not get very close to the screen or a very large
screen is needed. Rear projection is better when the users need
to get close to the screen. Rear projection also creates a
cleaner look.
Its important
to think about the users' natural distance from screen.
Its also important to think about throughput. Is this a device
where people will come and go quickly, say in a museum setting,
so it needs a big entryway and ways for people to enter and exit
at the same time, or will people be using the display for long
periods of time. To achieve the effect of imagery surrounding
the users it is usually necessary to have imagery on a door that
can be closed, but then it takes longer to get in and get out
Are there chairs and desks in the space? People can only stand
comfortably for so long.
Another increasingly important issue with displays and computers
is the power cost to operate the system. All of the components
are getting more power efficient over time but its important to
add up all of the power in use. Our current estimate is that
cave2 uses 50 kW, or about 16 times my home electricity usage if
the A/C is on at home, or about 200 times my home usage without
the A/C on.
Stereo Visuals
Stereo visuals, with each eye seeing a scene from a slightly
different point of view, can be very beneficial in certain
cases, and not beneficial at all in others. The stereo effect is
strongest within a few feet from your head and diminishes with
distance. Stereo visuals add a very strong sense of presence to
nearby objects, especially when coupled with head tracking. 3d
position in space can also be a very strong cue to help
disambiguate complex scenes.
With stereo imagery there are a variety of approaches. The most
common today are:
active stereo where
images are time multiplexed (left eye followed by right eye in
time and each eye of the glasses flicker in time)
passive stereo LCD
where the images are spatially multiplexed (left eye in the
first row of the image, right eye in the second row) with some
kind of polarized barrier strip in front of the LCD screen
matched to each eye of the glasses or having one
projector for the left eye image and one projector for the
right eye image with a different polarizing filter in front of
each projector)
autostereo - stereo
without glasses - is available on some small screens, and a
few TVs, but hasnt reached the quality level where it can be
deployed in large environments yet
Aligning stereo projectors
adds in more work, though the human visual system is pretty good
at fusing stereo images that are slightly misaligned.
Since passive glasses have no active components they are cheaper,
less fragile, and you don't have to replace the batteries
regularly
Another important feature with large curved or angled displays
is how stereo is computed. In virtual reality systems the
viewing position and the stereo visuals are typically computed
for the main tracked person - this means there is no stereo
separation looking 90 degrees to the right or right of the
tracked person and reversed stereo looking behind them since the
two images are computed for the tracked person's two eyes. For
larger groups in curved or angled displays its better to
generate the stereo for each screen assuming the user is looking
perpendicular to it so people can see stereo wherever they look.
Stereo visuals are
strongest when coupled to head-tracking. At the movie theatre
you see through the camera operators eyes. In tracked
environments you control the camera with your head motion giving
a very natural way to look around objects. Most trackers of this
kind have 6 degrees of freedom: motion in X, Y, Z and the
ability to sense Roll, Pitch, Yaw.
Tracking in the old
CAVE days made use of electromagnetic trackers, then acoustic
trackers + gyros, and now camera-based tracking. Electromagnetic
had the advantage of not needing line of site between the
emitters and the receivers, but was sucessiptible to
interference from metal objects and typically required a wired
connection between the user and the system. Camera based systems
are wireless, so they are more comfortable, but you need to have
enough cameras to cover the volume, and cameras can be occluded
when there are large numbers of people in the space
There are large displays; there are high-resolution displays;
there are large high-resolution displays; there are also small
high-resolution displays
The first small form factor ultra high resolution panel was the
IBM T220/T221 'big Bertha' that had 3820 x 2400 resolution in a
22" diagonal display at 204 ppi in 2001 for $18,000.
Today an iPad 4's 9.7" diagonal display has 2048 x 1536 resolution
at 264 ppi.
Today an HTC One phone's 4.7" diagonal screen has 1920 x 1080
pixels at 468 ppi.
The user is typically a foot away from these screens but could get
much closer to look at fine details.
Today a Macbook Pro's 15" diagonal display has 2880 x 1800
resolution at 220 ppi.
Today a google Chromebook Pixel's 13" diagonal display has 2560 x
1700 resolution at 239 ppi.
The user is typically two feet so away from these screens.
We
are at the point where our portable displays have more
resolution than we can see and the actual resolution numbers
will no longer matter. For large displays we sit back and look
at we are near that point. For large displays that we want to
walk up to and interact with, we are not there yet.
We are going to start thinking about displays in terms of whether
they give us 20/20 (6/6 in the rest of the world) at the natural
usage point.
The original cave had 8 ppi and you were typically standing 5'
from the screens
cave2 has 32 ppi and you are typically standing 8 to 10' from the
screens
now getting 4K LCD screens (3840 x 2160)
Sharp 32"
Panasonic 56"
Sony 55" and 65" and 84"
LG 84" $17,000
For
single person office-based high-resolution systems it can be
difficult to get projectors into that space, though there are
some nice small short throw distanace projectors out there now.
flat panels work more naturally in that confined space - e.g.
the tilings of displays used by the NSA or stock traders. There
are similar issues with resolution and border size. These tend
to be single user and single interaction device.
Networking
moving data or imagery to the computers
synchronizing the displays
Remote Rendering Resources
We have 100 gigabits coming into the lab and 80 of that going to
the cave.
Storage
local storage is often a mix of RAM, SSD Drives, traditional
drives, network storage
can duplicate a
(potentially really big) dataset on each node then each node
only has to access its local storage which minimizes network
traffic
can have each node
connect to the same shared drive over the network and use
local storage as a cache - reading pretty safe - writing can
lead to conflicts
can have the master
distribute data across the nodes as the application runs which
minimizes duplication of data files
Different applications have different local storage / local
network / remote storage / remote network needs that have to e
balanced.
Need to be sure there are reliable backups of the data and the
applications to be able to rebuild a node quickly if it fails, or
replace a drive that fails
GPUs
Increasing usage of GPUs to speed up and improve rendering - using
shaders to render data rather than creation of a large number of
polygons
Different GPUs have different abilities in synchronization
Best if image is
synchronized between the tlles
Best if there is no
tearing on individual tiles
GPU capabilities continue
to increase at a pleasant rate - still have issues getting data
onto the card, but that situation is also improving.
How many screens are driven by a single computer is an important
design decision.
Audio
Ambient sounds are useful to increase the believability of a space
Sounds are useful as a feedback mechanism
Important in collaborative applications to relay voice between the
various participants
Spatialized sound can be useful
Simple audio can be monaural with a single speaker but there are
many advantages to having multiple speakers as in a surround
sound system to give directional audio where sounds occur form a
particular location. As you get close to these sounds they get
louder. This can tell you where things have happened, or lead
you towards something.
In a surround environment such as cave2 sounds an alert you
to something happening out of your current field of view
ideally want a quiet environment with no fan noise from the
computers or displays so the only sounds you hear are those form
the application, or from your collaborators to minimize
distrations
Interaction
looking at the imagery
pointing at features / structures in the imagery by hand /
laser pointer / through a pointer on the screen
panning / zooming the imagery
turning elements on and off
changing time steps / playing / pausing of an animated
sequence
changing colour transfer function
changing data and processing upstream
moving closer to the screen to get a better view of details /
moving back to get the big picture
with head tracking being able to dynamically change the viewpoint
interaction with other devices - laptops, phones, tablets
We really want people to be 'looking through the display' at the
data ... that is the users stop talking about the display hardware
and spend their time talking about the content.
Some other papers relevant to our work here in evl:
The CAVE®: Audio Visual Experience Automatic Virtual Environment
Cruz-Neira, C., Sandin, D., DeFanti, T., Kenyon, R., Hart, J.
Communications of the ACM, 06/01/1992
http://www.evl.uic.edu/core.php?mod=4&type=3&indi=138
GeoWall: Stereoscopic Visualization for Geoscience Research and
Education
Johnson, A., Leigh, J., Morin, P., Van Keken, P.
IEEE Computer Graphics and Applications, 11/01/2006 - 12/31/2006
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4012558
Scalable resolution display walls J. Leigh, et. al. Proceedings of the IEEE 101, no. 1 (2013):
115–129.
http://www.evl.uic.edu/files/pdf/IEEE-ScalableWalls-May2012.pdf
CAVE2: A Hybrid Reality Environment for Immersive Simulation and
Information Analysis
Febretti, A., et. al.
Proceedings of IS&T/SPIE Electronic Imaging, The
Engineering Reality of Virtual Reality 2013, San Francisco, CA,
02/04/2013 - 02/05/2013
http://www.evl.uic.edu/files/pdf/SPIE13Paper-final.pdf
For Thursday's Class
If your UIN ends in an odd
number you should read this paper from 2000
Each person should produce a 1 page pdf file
critiquing the appropriate paper including your name, the
title of the paper, a short paragraph summary of the paper,
a list of things you found interesting in the paper,
and then what you think are the positives and negatives
about the research. In class on
Thursday everyone will use sage to drag and drop their file
onto the wall and a subset of the students will be asked to
talk more in depth about what you found in the paper
Coming Next Week
Making it Work -
Software
Before next class
please read
Ultrascale
Collaborative Visualization in Display-Rich Global
Cyberinfrastructure
Jeong, B., Leigh, J., Johnson, A., Renambot, L.,
Jagodic, R., Nam, S., Hur, H.
IEEE Computer Graphics and Applications Vol. 30.3 http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5453325
Again, you should produce a 1 page critique
of the paper to put up on the wall Tuesday in class, and a subset of the students will
be asked to talk more in depth about the paper.