Parallel Image Analysis

PhD-level course — 3 hp — period 4 (21 March – 27 May 2011) Centre for Image Analysis, Polacksbacken, Uppsala

Course Description

In this course we will explore techniques to take full advantage of modern desktop hardware, as well as high-performance computing facilities such as UPPMAX, for image analysis applications.

The course will give an introduction to:

how to use computing facilities to reduce the time it takes to run your long-running experiments;
how to turn a sequential algorithm into a parallel one;
how to use OpenMP to divide the work over all the cores in your multi-core computer;
how to use OpenCL and your graphics card to speed up simple algorithms;
how to use MPI for large-scale parallelism on distributed-memory computer clusters; and
how to use profiling tools to improve load balancing.

In addition to a simple exercise with each of these technologies, you will choose one of these technologies to try to speed up your favorite algorithm as much as you can. You will then share with the class your experiences and lessons learned. This project will serve as the course evaluation.

Goals

After the course, the student shall:

have some experience using a high-performance computing facility;
have implemented an image analysis algorithm that benefits from an increased number of computing cores;
have implemented an image filter that uses common graphics hardware;
have implemented an image analysis algorithm that can run on a distributed-memory computer cluster;
used a profiler; and
have translated a sequential algorithm to a parallel one.

Schedule

All lectures will be given in the CBA lecture room, Polacksbacken 2115.

1	Tuesday, 22 March, 13:15-15:00	Lecturer: Cris & Elias
	Introduction to image analysis, parallel computing, and UPPMAX. – Kalkyl user guide, UPPMAX file storage. – Slides.
	Wednesday, 23 March, 10:15-12:00	Lecturer: Elias
	Parallelism paradigms. – Slides.
	Wednesday, 23 March, 13:15-17:00	Hands-on session
	Using UPPMAX, scheduling jobs, etc. – Exercises.
2	Tuesday, 29 March, 13:15-15:00	Lecturer: Cris
	Taking advantage of your multi-core processor (pt 1). – OpenMP tutorial, OpenMP specification. – Slides (for both days). – Herb Sutter, "The Free Lunch Is Over".
	Wednesday, 30 March, 10:15-12:00	Lecturer: Cris
	Taking advantage of your multi-core processor (pt 2). – Németh and Palágyi, "Topology Preserving Parallel Thinning Algorithms." – Roerdink and Meijster, "The Watershed Transform: Definitions, Algorithms and Parallelization Strategies."
	Wednesday, 30 March, 13:15-17:00	Hands-on session
	OpenMP programming (on UPPMAX). – Exercises.
3	Tuesday, 5 April, 10:15-12:00	Lecturer: David
	Introduction to GPU architectures and introduction to OpenCL. David worked at Apple on the development of the first OpenCL implementation. – Slides: GPU Architectures for Non-Graphics People, Introduction to OpenCL.
	Wednesday, 6 April, 10:15-12:00	Lecturer: Cris
	Using your graphics card to speed up specific image analysis tasks. – The distance transform, erosion and separability, Meijster, Roerdink and Hesselink, "A General Algorithm for Computing Distance Transforms in Linear Time." – Xu and Mueller, "Real-time 3D computed tomographic reconstruction using commodity graphics hardware ."
	Wednesday, 6 April, 13:15-17:00	Hands-on session
	OpenCL programming (on laptop/desktop). – Exercises. – OpenCL specification, online manual pages.
4	Tuesday, 12 April, 13:15-15:00	Lecturer: Elias
	Using distributed-memory computer clusters (pt 1). – Slides (for both days).
	Wednesday, 13 April, 10:15-12:00	Lecturer: Elias
	Using distributed-memory computer clusters (pt 2).
	Wednesday, 13 April, 13:15-17:00	Hands-on session
	Message Passing Interface (MPI) programming, profiling MPI programs (on UPPMAX). – Exercises. – MPI-2.2 standard
5	Tuesday, 10 May, 13:15-17:00	Evaluation
	Project presentations by the students (15 minutes per group).

Information

Sign up for the course by emailing Cris Luengo. The deadline is one week before the course start. The course is limited to 24 participants, and taught in English. All lectures will be held at the Centre for Image Analysis.

Teachers:

Cris Luengo, CBA (responsible).
Elias Rudberg, UPPMAX.
David Black-Schaffer, Division of Computer Systems.

And support from:
Anders Hast, UPPMAX.

Prerequisites:

At least 5 hp in image analysis courses, and experience with MATLAB and C programming.

Equipment:

Students will need to use their own desktop/laptop for the project. Remote access to UPPMAX computing facilities will be given for the duration of the course.

Literature:

The course is based on journal papers, on-line software manuals, and so forth. Copies will be provided to the course participants during the course.