Conference presentation of a paper: Andrey Smorkalov, Mikhail Fominykh, and Mikhail Morozov: "Stream Processors Texture Generation Model for 3D Virtual Worlds: Learning Tools in vAcademia," in 9th International Symposium on Multimedia (ISM), Anaheim, CA, USA, December 9–11, 2013, IEEE.
Stream processors texture generation model for 3d virtual worlds learning tools in vacademia
1. Stream Processors Texture Generation
Model for 3D Virtual Worlds
Learning Tools in vAcademia
Andrey Smorkalov and Mikhail Morozov
Volga State University of Technology, Russia
Mikhail Fominykh
Norwegian University of Science and Technology, Norway
9th International Symposium on Multimedia (ISM)
December 9–11, 2013
Anaheim, CA, USA
1
VSUT
3. Motivation and challenges:
Applying 3D VWs for learning
o 3D Virtual Worlds (VWs)
– Have great features…
… but not widely used
o Challenges
– Steep learning curve
– Demand for computational and network resources
– lack of features that educators use in everyday teaching
o Solution Proposal
– Enabling learning scenarios which require large amounts
of 2D graphical content displayed
3
VSUT
4. Related work: Large Amount of
Graphics in 3D VWs
o Multiple workspaces or virtual screens
… but their performance is limited
o Small number of active screens (Second
Life has a limit of five)
o Static images (Sametime 3D has a sticky
notes tool, but notes are static, placed
on slots, constant size, and no other
tools on the same screen
o Individual use of screens
4
VSUT
7. Related work: Current
technological limitations
Usually, an image is calculated on a
CPU on client side (e.g., in Second
Life™ and Blue Mars™) or server side
(e.g., in Open Wonderland™) and then
loaded into the stream-processor
memory as a texture.
Therefore, the use of dynamic 2D
images in existing 3D VWs is very limited.
7
VSUT
11. Texture Generation Model:
Motivation
o CPU
‒ CPU is loaded maintaining 3D environment
‒ source data for the synthesis of images and the data area for the
resultant images are in the local memory of other devices
o Stream processors
‒ 3D visualization is hardware-based and conducted on SPs
‒ SPs’ computing power usually exceeds the capabilities of CPUs
tenfold
o Challenge
‒ SPs have hardware limitations which do not allow to use them for
implementing most of the classical image processing algorithms
11
VSUT
12. Texture Generation Model:
Mathematical Model (formalization)
o Defining
– Image, Transformation, Figure, Rasterization, Projected figure
o And configurable functionality
o texture sampling, color mask, hardware cut of the rasterization
area, hardware-based blending of the source image and the
rasterized image
o Calculating parts of image (even single
pixels instead of the whole image)
o Comparing the efficiency of different
approaches to any specific task
12
VSUT
13. Texture Generation Model:
Programming Model
The programming model and architecture
are based on four main objects
o Texture – image stored in SP memory
o Drawing Target defines resultant image
o Filter – subroutine returns color in coords.
o Filter Sequence – sequence of Filters
and limiting condition <β>
13
VSUT
14. Texture Generation Model:
Programming Model
o Modification of the DWT Algorithm
for SPs
‒
‒
Original modification of the Discrete Wavelet
Transformation (DWT) algorithm to run on SPs
We applied the method of 2D DWT filter cascade
o Rasterising Attributed Vector
Primitives on SPs
‒ SPs are able to deal only with vertexes and triangles
‒ We use a specific optimized method for triangulating
figures
14
VSUT
15. Original methods for processing
large amounts of graphics in 3D VWs
o Sharing Changing Blocks
‒ Sharing application window
‒ Sharing web-camera image
– Sharing video
– Sharing screen area
o Sharing Attributed Vector Figures
‒ Drawing figures and typing text
– Inserting text
o Processing Static Images
‒ Slideshow
‒ Image insert
‒ Sticky notes
15
– Area print screen
– Backchannel
VSUT
16. Original methods for processing
large amounts of graphics in 3D VWs
o Sharing Changing Blocks
‒ Sharing application window
‒ Sharing web-camera image
– Sharing video
– Sharing screen area
o Sharing Attributed Vector Figures
‒ Drawing figures and typing text
– Inserting text
o Processing Static Images
‒ Slideshow
‒ Image insert
‒ Sticky notes
16
– Area print screen
– Backchannel
VSUT
20. Performance Evaluation
I. Comparison of the algorithm
performance on SPs and CPU
II. General efficiency of the system
We present average results acquired by running the system on
‒ 20 different hardware configurations with Intel CPU and
NVidia / ATI graphics adapters from the same price range
‒ On each hardware configuration 10 runs were conducted for
each image size.
20
VSUT
21. Performance Evaluation:
I. Algorithms on SPs and CPU
The rationale behind using SPs (instead
of CPU) for image processing in
vAcademia is confirmed.
The improvement differs from the ratio
of the peaking performance of SPs to
the peaking performance of CPU not
more than twofold, which can be
considered satisfactory.
21
VSUT
22. Performance Evaluation:
II. General Efficiency of the System
Tested: performance degradation as a
function of the number of:
o VWBs (in one location)
o actively used VWBs
o simultaneous changes of images on
VWBs
22
VSUT
24. Performance degradation as a function
of the number of VWBs
Performance
100%
99%
98%
97%
Average
96%
Peaking
95%
94%
93%
92%
0
24
10
20
30
40
Number of whiteboards
50
VSUT
25. Performance degradation as a function
of the number of actively used VWBs
Performance
100%
95%
90%
Average
85%
Peaking
80%
75%
0
25
5
10
15
20
25
Number of actively used whiteboards
VSUT
26. Performance degradation as a function
of the number of simultaneous changes
of images on VWBs
Performance
100%
96%
92%
Average
88%
Peaking
84%
80%
1
26
2
3
4
5
Number of simultaneous changes of images
VSUT
27. User Evaluation
o Diagram designing task using
provided templates
o 23 second-year CS students
o No tutorials on vAcademia were
given
o All participants had experience
playing 3D video games
o Data: system logs, questionnaires,
and an interview
27
VSUT
29. User Evaluation
Question
It was clear what functions the VWB has and how to
access them.
It was comfortable "to look" at VWBs (to change the
view angle).
VWBs displayed the contents crisply and precisely
enough to understand them.
VWBs displayed the contents quickly enough, and
delays did not influence the process.
Increasing the # of VWBs in the virtual auditorium
during the class did not lead to visible delays.
VWB is a convenient (handy) enough tool for working
on similar tasks.
Working with vAcademia tools is more comfortable
than with traditional tools, for similar tasks.
It was clear how to work in vAcademia.
29
Str. agree Agree
16
7
15
8
14
9
14
8
13
10
13
8
15
8
19
4
N
VSUT
2
D
SD
30. Conclusions
o Original method for collaborative work
with large amount of graphical content
in 3D virtual worlds
o Design & implementation in vAcademia
o The algorithms we applied
– are superior to the commonly used ones
o The tools we designed
– have stable work and
– have educational value
30
VSUT
31. Future Work
o Designing scenarios for new learning
activities possible using our method
o Conducting a full-scale user
evaluation testing all designed tools
o Developing new tools based on our
method
31
VSUT