Department of Information and Computing Sciences

Departement Informatica Onderwijs
Bachelor Informatica Informatiekunde Kunstmatige intelligentie Master Computing Science Game&Media Technology Artifical Intelligence Human Computer Interaction Business Informatics

Onderwijs Informatica en Informatiekunde

Vak-informatie Informatica en Informatiekunde

Multimedia retrieval

Website:website containing additional information
Course code:INFOMR
Credits:7.5 ECTS
Period:period 1 (week 36 through 45, i.e., 2-9-2019 through 8-11-2019; retake week 2)
Timeslot:C
Participants:up till now 49 subscriptions
Schedule:Official schedule representation can be found in Osiris
Teachers:
formgrouptimeweekroomteacher
lecture   Tue 13.15-17.0037-44 BBG-214 Alex Telea
Thu 9.00-10.4536-44 BBG-214
Exam:
week: 45Tue 5-11-201913.30-16.30 uurroom: RUPPERT-BLAUW
week: 3Mon 13-1-202017.00-20.00 uurroom: RUPPERT-Bretake exam
Contents:

http://www.staff.science.uu.nl/~telea001/MR is the leading website and will be updated when needed. The current website (you are reading now) or the Osiris website cannot be assumed up-to-date and/or reliable.

Multimedia retrieval (MR) is about the search for and delivery of multimedia documents, such as text, images, video, audio, and 2D/3D shapes.

This course teaches MR from a bottom-up perspective. After introducing what MR is by means of examples and use-cases, the MR pipeline is presented. Next, each of the building blocks of this pipeline is discussed in detail, starting with the most basic one (data representation), going through the modeling of human perception of media, feature extraction, matching, evaluation, scalability, and presentation issue. At the end of the course, students should understand the theory, techniques, and tools that are involved in designing, building, and evaluating every block in the MR pipeline. The overall aim is thus for students to be able to design, build, and evaluate end-to-end MR systems for different types of multimedia data.

The course covers multimedia retrieval from a multidisciplinary perspective. Aspects taken into account: MR data representation; data (signal, image, shape) processing; understanding and working with high-dimensional data; connections between MR, machine learning, and data visualization; computational scalability and complexity aspects of working with big data collections; and human factors in interactive systems design.

The course takes a predominantly practical stance: After the theoretical principles of MR are introduced, we focus on how MR is to be practically implemented to be successful. Various design and implementation decisions for the MR pipeline building-blocks are discussed, focusing not only on their theoretical merits, but also ease of implementation/parameterization, robustness, and speed. Trade-offs between alternative solutions to a given problem are discussed.

Finally, as a 2nd year MSc course, this course has the meta goal to prepare students for their MSc graduation phase. This is done by teaching and assessing technical/scientific reporting and presentation skills.

Literature:

The course has no compulsory textbook, as a significant amount of information is presented in detail in its slides, papers, notes, and demos (all available online here). However, the following books are strongly recommended as optional reading material, as they give additional details on the material discussed in the course:

  • Handbook of Multimedia Information Retrieval (H. Eidenberger; publisher: Atpress; publication date: 2012; index information: ISBN 9783848222834)
  • Shape Analysis and Classification: Theory and Practice (L. Da Fontoura Costa, R. Marcondes Cesar Jr.; publisher: CRC Press; publication date: 2001 (subsequent editions are also fine); index information: ISBN 9780849334931 (for 1st edition))
  • Data Visualization - Principles and Practice (2nd edition) (A. C. Telea; publisher: CRC Press; publication date: 2014; index information: ISBN 9781466585263)

Visit the course page to find out which chapters from the above books cover which topics of the course.

Course form:The course consists of lectures, self-study, presentations, and a project.
Exam form:The course consists of lectures, self-study/work groups, and a project. The project-based assessment reflects the practical nature of MR: Students are asked to design and end-to-end MR pipeline, comment on their design choices, evaluate the pipeline, and comment on the strengths and weaknesses of the observed results. The project is assessed by means of weekly updates (submitted by the students, assessed by the lecturer); a final oral presentation (including a demo); and a final technical report (covering all aspects of the work done to solve the problem).

Our experience learned us that genuine active participation is needed to pass the course. Calculation grade:

  • 25% Process (consistency, quality, and completeness of the weekly-submitted project updates)
  • 25% Final project presentation
  • 50% Project report

Each of the above three elements is graded separately. The final grade is the weighted average of these three grades.

To pass the course, the final grade has to be at least 5.

Minimum effort to qualify for 2nd chance exam:To be admitted to the 2nd chance exam (aanvullende toets) the first exam result (oorspronkelijke uitslag) should be minimally a 4.
Description:This course has objectives to train students in:
  1. General concepts of multimedia retrieval (data types, multimedia human perception, feature extraction, distances and matching, evaluation, scalability, and visual presentation aspects)
  2. Multimedia data processing (signal processing, 2D image processing, 3D shape processing)
  3. Conducting scientific research, as preparation for your MSc-project: understanding and refining a research question; proposing a research plan, with milestones; executing the research plan, possibly adapting the initial design ideas; evaluate objectively, quantitatively and qualitatively, the obtained results; presenting the results both orally and in a scientific report (with due citations for replicabiility).
As such, this course is complementary to other courses, in particular data-analyse en retrieval, pattern recognition and computer vision.
wijzigen?