Course detail

Computer Vision (in English)

FIT-POVaAcad. year: 2026/2027

Introduction to the principles and methods of computer vision: image formation and scene geometry, image features and their matching, geometric scene reconstruction, motion analysis and localization, recognition and segmentation of visual content, learning of visual representations, connecting vision and language, 3D computer vision methods, video analysis, and open problems in computer vision.

Language of instruction

English

Number of ECTS credits

Assignment to study programme types

Master's

Mode of study

Not applicable.

Guarantor

prof. Dr. Ing. Pavel Zemčík, dr. h. c.

Department

Department of Computer Graphics and Multimedia (UPGM)

Offered to foreign students

Of all faculties

Entry knowledge

Not applicable.

Rules for evaluation and completion of the course

Two home assignments, mid-term test and individual project.

Aims

Students will gain a comprehensive overview of the principles and methods of computer vision, ranging from geometric approaches to current methods based on neural networks. They will understand the process of scene capture, camera calibration and the reconstruction of 3D information, and will apply this knowledge in practice in an individual homework assignment. They will become familiar with representations of image data and with modern approaches to object detection and recognition. In a practical assignment, they will build their own dataset and train their own object detector. They will further apply the acquired knowledge in a team project on a selected topic. At the same time, they will improve their skills in working with image processing and machine learning tools in Python, in data preparation and annotation, and in the evaluation of experimental results.

Study aids

Not applicable.

Prerequisites and corequisites

Not applicable.

Basic literature

Not applicable.

Type of course unit

Lecture

26 hours, optionally

Teacher / Lecturer

Ing. Michal Hradiš, Ph.D.
Ing. Marek Šolony, Ph.D.
Ing. Marek Vaško
prof. Dr. Ing. Pavel Zemčík, dr. h. c.
doc. Ing. Vítězslav Beran, Ph.D.
doc. Ing. Michal Španěl, Ph.D.
prof. Ing. Martin Čadík, Ph.D.

Syllabus

Introduction, motivation and applications. (Zemčík 15.9.)
Image formation and cameras. Projection, camera calibration, distortion and coordinate systems. (Zemčík 22.9.)
Features and correspondence. Classical detectors and descriptors (Harris, SIFT), learned features and matchers. (Beran 29.9.)
Robust geometric estimation. Homographies, RANSAC, model fitting, single camera. (Čadík 6.10.)
Two-view and multi-view geometry. Epipolar geometry, triangulation, and structure from motion (SfM). (Šolony, 13.10.)
Motion and localization. Bundle adjustment, state tracking, visual odometry, and SLAM. (Šolony, 20.10.)
Neural networks for vision. From classical features to convolutional neural networks (CNNs) and vision transformers (ViTs). (Hradiš, 27.10.)
Optical flow and dense matching. Neural optical flow, sparse and dense matching, and an outlook on end-to-end 3D reconstruction. (Čadík 3.11.)
(Test) Depth and modern 3D. Monocular depth estimation, point clouds, volumetric representations, and an overview of neural fields (NeRF) and Gaussian splatting. (Španěl 10.11.)
Public holiday (17.11.)
Detection and segmentation. From anchor and window-based formulations to YOLO and DETR; semantic and instance segmentation; SAM-style promptable models. (Vaško 24.11.)
Visual representation learning and vision–language models. Transfer learning, self-supervised learning, and DINO-style representations; CLIP, open-vocabulary recognition, visual grounding, and vision–language models (VLMs). (Hradiš, 1.12.)
Current trends and embedded vision. State-of-the-art demonstrations and efficient deployment of vision models on embedded devices. (8.12.)

NOTE: The topics are not guaranteed exactly, and will be continuously updated.

Project

26 hours, compulsory

Teacher / Lecturer

Ing. Martin Kostelník
Ing. Michal Hradiš, Ph.D.

Syllabus

The project work in the course consists of two parts:

(10 pts) Two individual homework assignments. In the first half of the semester, students calibrate a camera and project a 3D animation into real video (augmented reality). In the second half of the semester, they create their own dataset and train their own object detector.
(30 pts) A team project for 2–3 students on selected topics. After consultation, students may propose a topic of their own. The results will be presented during a poster session at the end of the semester. Projects from past years can be seen here: https://www.fit.vut.cz/person/ikostelnik/public/knn-pova-projects/.

Elearning

eLearning: currently opened course