Computer Vision Methods

Schedule on FEL (CZ course) Schedule on FEL (EN course)
Upload system
Discussion forum Discussion forum archive (2022/2023)
Labs

Course Description

This course focuses on the following computer vision problems: finding correspondences between images using image features and their robust invariant descriptors, image retrieval, object detection and recognition, and visual tracking.

Pre-requisites

The course has no formal pre-requisites. However, certain skills and knowledge are assumed, and it is the responsibility of the student to get to the required level.

The assignments are implemented in the Python, numpy, pytorch computing environment, mostly in form of jupyter notebooks, and familiarity with it will help. The programing assignments, involving either implementing, modifying or testing computer vision methods, are a substantial part of the labs.

Knowledge of the basics of digital image processing as convolution, filtration, intensity transformations, image function interpolations and basic geometric transformations of the image (see the first lab) is assumed. Knowledge of linear algebra and probability theory is needed to understand the presented computer vision methods.

Lectures: Monday 9:15 - 10:45, KN:E-107

Lecturers: JM Jiří Matas, JC Jan Čech, DM Dmytro Mishkin, GT Giorgos Tolias, MS Milan Šulc

Note: some of the lectures have changed, but the 2021 recordings mostly provide a good idea about the content.
Lectures will be streamed on YouTube, live link: https://www.youtube.com/playlist?list=PLQL6z4JeTTQkqF6KkcZZDi2KFwky9SQpq
Recorded lectures - playlist: https://www.youtube.com/playlist?list=PLQL6z4JeTTQnozHfghnzq3AK-lIBZmQdP
Recorded lectures - playlist (2023): https://www.youtube.com/playlist?list=PLQL6z4JeTTQneuiXekoB639gEOzuen79_

Changes compared to the last year are in italics.

Week	Date	Lecturer	Slides	Topic
1	19.2.	JM	Slides 2023 ( 1st lecture slides 1-65), recording 2023,recording 2021	Correspondences and wide baseline stereo. Motivation and applications. Intro in the image processing. Perspective pinhole camera model. Interest point and distinguished regions detection: Moravec detector (corner detection)
2	26.2.	DM	Slides 2023 ( 2nd lecture slides 65--178), Slides 2022, recording 2023 recording 2021	Harris operator. Image gradient. Laplace operator and its approximation by difference of Gaussians, Hessian detector, affine covariant version. Descriptors of measurement regions: SIFT (scale invariant feature transform), RootSIFT.
3	4.3.	JM	RANSAC recording 2021	RANSAC.
4	11.3.	DM	Slides 2023 ( 3rd lecture slides: 179 -- end) Slides 2022, recording 2021	Matching. Deep learned features: R2D2, SuperGlue.
5	18.3.	GT	Image Retrieval recording	task formulation, evaluation metrics, Bag-of-Words, VLAD, spatial verification, special objectives: zoom in/out .
6	25.3.	JC	Deep learning recording 2024	A shallow introduction into the deep machine learning. Convolutional Neural Networks, Transformers. Principles, layers, architectures for image recognition.
7	1.4.		Easter Monday
8	8.4.	JC	Deep learning II recording 2024	Deep architectures for object detection and semantic segmentation. Further insights into the deep nets. Foundation models (CLIP, DINO, Segment Anything, Depth Anything).
9	15.4.	GT	Deep Metric Learning recording	architectures, losses, types of supervision
10	22.4.	GT	Self-supervised Representation Learning recording	tasks with self-supervisory signal, auto-encoders, learning via augmentations, contrastive approaches
11	29.4.	JM	KLT, recording 2021	Tracking I. Introduction. Kanade-Lucas-Tomasi tracker.
12	6.5.	JM	KCF Tracking TLD, Tracking_by_Segmentation recording 2021	Tracking II. KCF Kernel Correlation Filter. Long-term Tracking, TLD: Tracking-Learning-Detection, Tracking by Segmentation. Introduction to KCF lab task.
13	13.5.	JM	Hough Transform recording 2021	Hough transform .
14	20.5.	JC	Deep Generative Models	Generative modelling for Computer Vision

Evaluation

Work during the semester 50%, written part of the exam 40%, oral part of the exam 10%.

Note that the points from the labs are reweighted using a “normalization factor” so that they correspond to 50% of your evaluation. That means, the points from the labs that contribute to your exam, are (your total number of points from semester including bonus points)/(sum of points available from all non-bonus tasks) * 50.

Exam

Examples of exam questions. There will be 4-5 similar questions at the written part of the exam. The oral part of the questions takes place after the written part and will focused on discussion of your answers.

Literature

Lecture slides constitute the main source of study literature in this course.

Further Info

Further information is available in next sections of this page. We would appreciate your feedback on the contents and organization on the discussion forum of the course.

Good luck to all participants of the course.

Lecturers

Jiří Matas	Jan Čech	Giorgos Tolias	Dmytro Mishkin

Consultations are possible upon request.

Table of Contents