Computer Vision – Theory and Practice

Course guarantor:

Ondřej Chum

Lecturer:

Ondřej Chum

Tutor:

Ondřej Chum

Supervisor:

Department of Cybernetics

Synopsis:

In the course, the PhD students will study selected sophisticated state-of-the-art computer-vision methods that have an efficient implementation publically available. The course will focus on general methods that have been successfully used in a number of applications, including large scale search in high-dimensional spaces, deep neural networks, and the graph labelling algorithms. The methods selected for the course evolve based on the current progress in the field; the selection is also alternated by the student’s interests. The goal for the students is to understand the method, to understand the implementation, and to be able to use the implementation as a tool to solve other problems.

Requirements:

Syllabus of lectures:

The course will include two strands. The first strand will be similar to a reading group - the students will individually study a self-contained material, typically a published paper. The methods are then discussed at the lecture. In the second, practical component of the course, the students will use an implementation of the discussed methods to solve some particular task. The solutions and their properties are also discussed at the lecture.

Students are expected to have a basic knowledge of algorithms and data structures, and the ability of individual work. The scope of the course is suitable for PhD students as a preparation for scientific work, and is not suitable for common undergraduate students.

An example of studied methods:

1. Marius Muja and David G. Lowe: “Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration”, in International Conference on Computer Vision Theory and Applications (VISAPP'09), 2009

2. Herve Jegou, Matthijs Douze, Cordelia Schmid: “Product quantization for nearest neighbor search”, PAMI 2011.

3. Wei Dong, Moses Charikar, Kai Li : “Efficient K-Nearest Neighbor Graph Construction for Generic Similarity Measures.” In Proc. of the International conference on World Wide Web (WWW). New York, NY. 2011.

4. Jeff Johnson, Matthijs Douze, Hervé Jégou: “Billion-scale similarity search with GPUs” 2017

Due to a high demand on the individual work of the students, the course is assessed by 4 credits.

Syllabus of tutorials:

Study Objective:

Study materials:

List of bibliography available on course web-pages.

Note:

Time-table for winter semester 2024/2025:

Time-table is not available yet

Time-table for summer semester 2024/2025:

Time-table is not available yet

The course is a part of the following study plans:

• Doctoral studies, daily studies (compulsory elective course)
• Doctoral studies, combined studies (compulsory elective course)