SD-VBS: The san diego vision benchmark suite
Research output: Contribution to journal › Conference article › Research › peer-review
Standard
SD-VBS : The san diego vision benchmark suite. / Venkata, Sravanthi Kota; Ahn, Ikkjin; Jeon, Donghwan; Gupta, Anshuman; Louie, Christopher; Garcia, Saturnino; Belongie, Serge; Taylor, Michael Bedford.
In: Proceedings of the 2009 IEEE International Symposium on Workload Characterization, IISWC 2009, 2009, p. 55-64.Research output: Contribution to journal › Conference article › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - SD-VBS
T2 - 2009 IEEE International Symposium on Workload Characterization, IISWC 2009
AU - Venkata, Sravanthi Kota
AU - Ahn, Ikkjin
AU - Jeon, Donghwan
AU - Gupta, Anshuman
AU - Louie, Christopher
AU - Garcia, Saturnino
AU - Belongie, Serge
AU - Taylor, Michael Bedford
PY - 2009
Y1 - 2009
N2 - In the era of multi-core, computer vision has emerged as an exciting application area which promises to continue to drive the demand for both more powerful and more energy efficient processors. Although there is still a long way to go, vision has matured significantly over the last few decades, and the list of applications that are useful to end users continues to grow. The parallelism inherent in vision applications makes them a promising workload for multi-core and many-core processors. While the vision community has focused many years on improving the accuracy of vision algorithms, a major barrier to the study of their computational properties has been the lack of a benchmark suite that simultaneously spans a wide portion of the vision space and is accessible in a portable form that the architecture community can easily use. We present the San Diego Vision Benchmark Suite (SD-VBS), a suite of diverse vision applications drawn from the vision domain. The applications are drawn from the current state-of-the-art in computer vision, in consultation with vision researchers. Each benchmark is provided in both MATLAB and C form. MATLAB is the preferred language of vision researchers, while C makes it easier to map the applications to research platforms. The C code minimizes pointer usage and employs clean constructs to make them easier for parallelization. Furthermore, we provide a spectrum of input sets that enable researchers to control simulation time, and to understand properties as inputs increase to leverage better processor performance. In this paper, we describe the benchmarks, show how their runtime is attributed to their constituent kernels, overview some of their computational properties - including parallelism - and show how they are affected by growing inputs. The benchmark suite will be made available on the Internet, and updated as new applications emerge.
AB - In the era of multi-core, computer vision has emerged as an exciting application area which promises to continue to drive the demand for both more powerful and more energy efficient processors. Although there is still a long way to go, vision has matured significantly over the last few decades, and the list of applications that are useful to end users continues to grow. The parallelism inherent in vision applications makes them a promising workload for multi-core and many-core processors. While the vision community has focused many years on improving the accuracy of vision algorithms, a major barrier to the study of their computational properties has been the lack of a benchmark suite that simultaneously spans a wide portion of the vision space and is accessible in a portable form that the architecture community can easily use. We present the San Diego Vision Benchmark Suite (SD-VBS), a suite of diverse vision applications drawn from the vision domain. The applications are drawn from the current state-of-the-art in computer vision, in consultation with vision researchers. Each benchmark is provided in both MATLAB and C form. MATLAB is the preferred language of vision researchers, while C makes it easier to map the applications to research platforms. The C code minimizes pointer usage and employs clean constructs to make them easier for parallelization. Furthermore, we provide a spectrum of input sets that enable researchers to control simulation time, and to understand properties as inputs increase to leverage better processor performance. In this paper, we describe the benchmarks, show how their runtime is attributed to their constituent kernels, overview some of their computational properties - including parallelism - and show how they are affected by growing inputs. The benchmark suite will be made available on the Internet, and updated as new applications emerge.
UR - http://www.scopus.com/inward/record.url?scp=70649096324&partnerID=8YFLogxK
U2 - 10.1109/IISWC.2009.5306794
DO - 10.1109/IISWC.2009.5306794
M3 - Conference article
AN - SCOPUS:70649096324
SP - 55
EP - 64
JO - Proceedings of the 2009 IEEE International Symposium on Workload Characterization, IISWC 2009
JF - Proceedings of the 2009 IEEE International Symposium on Workload Characterization, IISWC 2009
Y2 - 4 October 2009 through 6 October 2009
ER -
ID: 302050093