Computer Vision Methods

The course covers selected computer vision problems: search for correspondences between images via interest point detection, description and matching, image stitching, detection, recognition and segmentation of objects in images and videos, image retrieval from large databases and tracking of objects in video sequences.

This course is also part of the inter-university programme prg.ai Minor. It pools the best of AI education in Prague to provide students with a deeper and broader insight into the field of artificial intelligence. More information is available at https://prg.ai/minor.

Požadavky:

Knowledge of calculus and linear algebra.

Osnova přednášek:

1. Correspondences and wide baseline stereo I. Motivation and applications. Perspective pinhole camera model.

Interest point and distinguished regions detection: Harris detector (corner detection)

2.Correspondences and wide baseline stereo II. Laplace operator and its approximation by difference of Gaussians. Affine covariant version. Descriptors of SIFT (scale invariant feature transform), RootSIFT.

Multiview feature matching. Deep learned features: R2D2, Super Glue.

3.RANSAC (Random Sample and Consensus)

4.3D reconstruction I.

5.3D reconstruction II

6.Deep learning I. Convolutional Neural Networks, Transformers. Architectures for image recognition.

7.Deep learning II. Architectures for object detection and semantic segmentation.. Foundation models (CLIP, DINO, Segment Anything, Depth Anything)

8.Tracking I. Problem formulation. KLT - Lucas-Kanade tracker, DCF - discriminative correlation tracker.

9.Tracking II. Long-term tracking.

10.Image Retrieval I., Bag-of-Words, VLAD, spatial verification, special objectives: zoom in/out .

11.Image Retrieval II. deep metric learning, architectures, losses

12.Self-supervised representation learning. Auto-encoders, learning via augmentations, contrastive approaches

13.Generative modelling for Computer Vision

Osnova cvičení:

1.Introduction to Image Processing in python using PyTorch.

2.Debugging pytorch.

3.Correspondence problem I, detection of the interest points.

4.Correspondence problem II, computing local invariant description.

5.Correspondence problem III, finding tenative correspondences and RANSAC.

6.Correspondence problem, summary.

7.Convolutional Neural Networks: training a classifier.D LN recording 2022

8.Convolutional Neural Networks II: debugging training process.

9.Image Retrieval, BoW TF-IDF, fast spatial verification.

10.Assignment defence.

11.Deep metric learning.

12.Self-supervised Learning.

13.Tracking.

Cíle studia:

The methods for image registration, retrieval and for object detection and tracking are explained. In the labs, the students implement selected methods and test performance on real-world problems.

Studijní materiály:

D. A. Forsyth, J. Ponce. Computer Vision: A Modern Approach. Prentice Hall 2003

I. Goodfellow, Y. Bengio, A. Courville: Deep Learning, 2016

A. Torralba, P. Isola, W. T. Freeman: Foundations of Computer Vision,

Poznámka:

URL: https://cw.fel.cvut.cz/wiki/courses/mpv/start

Další informace:

https://cw.fel.cvut.cz/wiki/courses/mpv/start

Pro tento předmět se rozvrh nepřipravuje

Předmět je součástí následujících studijních plánů:

Open Informatics - Computer Vision and Image Processing (povinný předmět oboru)
Medical Electronics and Bioinformatics - Specialization Image Processing (PS)
Medical Electronics and Bioinformatics - Specialization Signal Processing (povinně volitelný předmět)
Medical Electronics and Bioinformatics - Specialization Bioinformatics (povinně volitelný předmět)
Medical Electronics and Bioinformatics - Specialization Medical Instrumentation (povinně volitelný předmět)
Cybernetics and Robotics (povinně volitelný předmět)
prg.ai Master (povinný předmět programu)
Open Informatics - Autonomous Systems 2026 (PS)
Open Informatics - Artificial Intelligence 2026 (povinně volitelný předmět)
Open Informatics - Computer Vision 2026 (PS)