Presentation is loading. Please wait.

Presentation is loading. Please wait.

Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Patrick Eibl, Dennis Healy, Nikos P.

Published byPenelope Hancock Modified over 9 years ago

Similar presentations

Presentation on theme: "Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Patrick Eibl, Dennis Healy, Nikos P."— Presentation transcript:

1 Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Patrick Eibl, Dennis Healy, Nikos P. Pitsianis and Xiaobai Sun HPEC 2009 MIT Lincoln Lab

2 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS In Memory of Dennis M. Healy, Jr. Sept 23 2009HPEC 2009 2 July 2007 Contributed by Mary Healy Hambric

3 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS LCCs : local correlation coefficients –application requirements –computational challenges –recent progress Fast LCC –Algorithmic acceleration –Architectural acceleration (on GPUs) Building blocks –In image processing and information processing Conclusion Outline Sept 23 2009HPEC 2009 3

4 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Fast Convolution or Correlation Sept 23 2009HPEC 2009 4 Discrete Version Equally spaced sampling Periodic convolution : padding otherwise FFT : O( N log(N) ) Convolution (Correlation) Theorem

5 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Calculating the LCCs of a prototype-pattern with all patches in a large image  A basic operation in image registration, object classification and recognition, target identification and tracking  Statistically robust to local changes and noise Fast calculation of LCCs without compromising integrity  in many situations, rapid computation of LCCs was based on relaxed linearized conditions, to avoid local translation in mean and normalization in variance  “Fast LCC with local zero-mean translation and unit-variance normalization”, Sun & Pitsianis 2008 Local Correlation Coefficients (LCCs) Sept 23 2009HPEC 2009 5

6 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Applications: MONTAGE Sept 23 2009HPEC 2009 6 A. Portnoy, et al., Design and characterization of thin multiple aperture infrared cameras Applied Optics, Vol. 48 Issue 11, pp.2115-2126 (2009)Design and characterization of thin multiple aperture infrared cameras A. Wagadarikar, et al., Video rate spectral imaging using a coded aperture snapshot spectral imager Optics Express, Vol. 17 Issue 8, pp.6368-6388 (2009)Video rate spectral imaging using a coded aperture snapshot spectral imager M. Shankar, et al., Thin infrared imaging systems through multichannel sampling Applied Optics, 47(10):B1-B10 (January 2008)Thin infrared imaging systems through multichannel sampling N. P. Pitsianis, et al, Compressive imaging sensors, In Proc. SPIE Vol. 6232, Intelligent Integrated Microsystems; Ravindra A. Athale, John C. Zolper; Eds. pp 43-51, May 2006Compressive imaging sensors

7 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Sub-pixel Image Registration Sept 23 2009 7 HPEC 2009

8 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Template Matching Sept 23 2009HPEC 2009 8

9 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Without Local Normalization Sept 23 2009HPEC 2009 9

10 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS With Local Normalization Sept 23 2009HPEC 2009 10

11 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Local Correlation Coefficients Sept 23 2009HPEC 2009 11 PT S T P

12 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Normalized Local Correlation Coefficients Sept 23 2009HPEC 2009 12 X. Sun, N. P. Pitsianis, and P. Bientinesi, Fast computation of local correlation coefficients Proc. SPIE 7074, 707405 (2008)Fast computation of local correlation coefficients PT S T P

13 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Sept 23 2009HPEC 2009 13 Algorithmic Acceleration 1.Reformulation in term of convolutions 2.Use of Fourier transform identities

14 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Reformulation in Terms of Convolutions Sept 23 2009HPEC 2009 14

15 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Fast LCC in the Fourier Domain Sept 23 2009HPEC 2009 15 2.5 FFTs per image

16 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS DFT Identity with real data : Two by one Sept 23 2009HPEC 2009 16

17 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Sept 23 2009HPEC 2009 17 Illustration of Two-by-One identity

18 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS DFT Identity : One-by-Half Sept 23 2009HPEC 2009 18

19 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Sept 23 2009HPEC 2009 19 Illustration of One-by-Half + -

20 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Sept 23 2009HPEC 2009 20 Architectural Acceleration 1.Necessary complement of fast algorithms in real-time image processing or massive data processing 2.The use of GPUs in particular  omnipresent in desktops, laptops  a prototype of parallel architecture  many-core  memory hierarchy  bandwidth  programming model  3D image processing

21 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS 2D Convolution on CPU and GPU Sept 23 2009HPEC 2009 21 FFTW 3.1.2 on Xeon CUDA SDK 2.3 CUFFT on GPU Execution time (msec)

22 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Algorithmic Acceleration of 2D Convolution Sept 23 2009HPEC 2009 22

23 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Algorithmic Acceleration of 3D Convolution Sept 23 2009HPEC 2009 23

24 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS GPU Acceleration of 2D LCCs Sept 23 2009HPEC 2009 24 Streamed FLCC 4Kx4K FLCC 2Kx2K Streamed FLCC 2Kx2K FLCC 4Kx4K

25 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS GPU Acceleration of 3D LCCs Sept 23 2009HPEC 2009 25 Streamed FLCC FLCC 128x128x128 Streamed FLCC FLCC 256x256x256

26 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Sept 23 2009HPEC 2009 26 Building Blocks 1.In image processing 2.In data modeling & information processing

27 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Construction of spectral images from coded intensity data Fast Deconvolution with Incomplete Data Sept 23 2009HPEC 2009 27 X. Sun and N. P. Pitsianis (2008)

28 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Object Feature Extraction in Multiple Layers  Simple and densely populated features at lower layers  Complex but sparse features at higher layers Hierarchical Data Indexing & Fast Feature Matching Sept 23 2009HPEC 2009 28 S. Fidler, G. Berginc, A. Leonardis, Proc IEEE CVPR (2006)

29 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Convolution/correlation, LCCs important in new and conventional applications Reformulation critical in designing fast algorithms without compromising quality Joint algorithmic and architectural accelerations Efficient implementation of fast algorithms on modern parallel architectures Conclusions Sept 23 2009HPEC 2009 29

30 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS FANTOM : Algorithm-Architecture Co-design –Supported in part by DARPA MTO AutoGPU –Supported in part by MC-FP7 –Donation of GPUs by NVIDIA CCF-FMM –Supported in part by NSF G. Papamakarios and G. Rizos @ Aristotle Univ. Acknowledgments Sept 23 2009HPEC 2009 30

Download ppt "Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 FAST PATTERN MATCHING IN 3D IMAGES ON GPUS Patrick Eibl, Dennis Healy, Nikos P."

Similar presentations

Image Registration  Mapping of Evolution. Registration Goals Assume the correspondences are known Find such f() and g() such that the images are best.

Image Registration  Mapping of Evolution. Registration Goals Assume the correspondences are known Find such f() and g() such that the images are best.
Compressed sensing Carlos Becker, Guillaume Lemaître & Peter Rennert

Compressed sensing Carlos Becker, Guillaume Lemaître & Peter Rennert
DUKE CSNUFFT on Multicores HPEC-2008 Sept. 25, 2008NUFFTs on Multicores 1 Parallelization of NUFFT with Radial Data on Multicore Processors Nikos Pitsianis.

DUKE CSNUFFT on Multicores HPEC-2008 Sept. 25, 2008NUFFTs on Multicores 1 Parallelization of NUFFT with Radial Data on Multicore Processors Nikos Pitsianis.
A Matlab Playground for JPEG Andy Pekarske Nikolay Kolev.

A Matlab Playground for JPEG Andy Pekarske Nikolay Kolev.
ECE 562 Computer Architecture and Design Project: Improving Feature Extraction Using SIFT on GPU Rodrigo Savage, Wo-Tak Wu.

ECE 562 Computer Architecture and Design Project: Improving Feature Extraction Using SIFT on GPU Rodrigo Savage, Wo-Tak Wu.
CUDA Programming Lei Zhou, Yafeng Yin, Yanzhi Ren, Hong Man, Yingying Chen.

CUDA Programming Lei Zhou, Yafeng Yin, Yanzhi Ren, Hong Man, Yingying Chen.
Acceleration on many-cores CPUs and GPUs Dinesh Manocha Lauri Savioja.

Acceleration on many-cores CPUs and GPUs Dinesh Manocha Lauri Savioja.
Spectral Processing of Point-sampled Geometry

Spectral Processing of Point-sampled Geometry
A Performance and Energy Comparison of FPGAs, GPUs, and Multicores for Sliding-Window Applications From J. Fowers, G. Brown, P. Cooke, and G. Stitt, University.

A Performance and Energy Comparison of FPGAs, GPUs, and Multicores for Sliding-Window Applications From J. Fowers, G. Brown, P. Cooke, and G. Stitt, University.
REALTIME OBJECT-OF-INTEREST TRACKING BY LEARNING COMPOSITE PATCH-BASED TEMPLATES Yuanlu Xu, Hongfei Zhou, Qing Wang*, Liang Lin Sun Yat-sen University,

REALTIME OBJECT-OF-INTEREST TRACKING BY LEARNING COMPOSITE PATCH-BASED TEMPLATES Yuanlu Xu, Hongfei Zhou, Qing Wang*, Liang Lin Sun Yat-sen University,
GPGPU platforms GP - General Purpose computation using GPU

GPGPU platforms GP - General Purpose computation using GPU
Synergy.cs.vt.edu Power and Performance Characterization of Computational Kernels on the GPU Yang Jiao, Heshan Lin, Pavan Balaji (ANL), Wu-chun Feng.

Synergy.cs.vt.edu Power and Performance Characterization of Computational Kernels on the GPU Yang Jiao, Heshan Lin, Pavan Balaji (ANL), Wu-chun Feng.
Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 Fast Extraction of Feature Salience Maps for Rapid Video Data Analysis Nikos P.

Computer Science Department, Duke UniversityPhD Defense TalkMay 4, 2005 Fast Extraction of Feature Salience Maps for Rapid Video Data Analysis Nikos P.
BENCHMARK SUITE RADAR SIGNAL & DATA PROCESSING CERES EPC WORKSHOP 2008-10-01.

BENCHMARK SUITE RADAR SIGNAL & DATA PROCESSING CERES EPC WORKSHOP
Phase Retrieval Applied to Asteroid Silhouette Characterization by Stellar Occultation Russell Trahan & David Hyland JPL Foundry Meeting – April 21, 2014.

Phase Retrieval Applied to Asteroid Silhouette Characterization by Stellar Occultation Russell Trahan & David Hyland JPL Foundry Meeting – April 21, 2014.
Computationally Efficient Histopathological Image Analysis: Use of GPUs for Classification of Stromal Development Olcay Sertel 1,2, Antonio Ruiz 3, Umit.

Computationally Efficient Histopathological Image Analysis: Use of GPUs for Classification of Stromal Development Olcay Sertel 1,2, Antonio Ruiz 3, Umit.
Christopher Mitchell CDA 6938, Spring 2009. The Discrete Cosine Transform  In the same family as the Fourier Transform  Converts data to frequency domain.

Christopher Mitchell CDA 6938, Spring The Discrete Cosine Transform  In the same family as the Fourier Transform  Converts data to frequency domain.
Object Based Video Coding - A Multimedia Communication Perspective Muhammad Hassan Khan 2004-03-0020.

Object Based Video Coding - A Multimedia Communication Perspective Muhammad Hassan Khan
By Arun Bhandari Course: HPC Date: 01/28/12. GPU (Graphics Processing Unit) High performance many core processors Only used to accelerate certain parts.

By Arun Bhandari Course: HPC Date: 01/28/12. GPU (Graphics Processing Unit) High performance many core processors Only used to accelerate certain parts.
Parallel Applications Parallel Hardware Parallel Software IT industry (Silicon Valley) Users Efficient Parallel CKY Parsing on GPUs Youngmin Yi (University.

Parallel Applications Parallel Hardware Parallel Software IT industry (Silicon Valley) Users Efficient Parallel CKY Parsing on GPUs Youngmin Yi (University.

Similar presentations

Ads by Google