• A
  • A
  • A
  • АБB
  • АБB
  • АБB
  • А
  • А
  • А
  • А
  • А
Обычная версия сайта
Книга
Integral Robot Technologies and Speech Behavior

Kharlamov A. A., Pantiukhin D., Borisov V. et al.

Newcastle upon Tyne: Cambridge Scholars Publishing, 2024.

Статья
Clique detection with a given reliability

Semenov D., Koldanov A. P., Koldanov P. et al.

Annals of Mathematics and Artificial Intelligence. 2024.

Глава в книге
Neural Networks for Speech Synthesis of Voice Assistants and Singing Machines

Pantiukhin D.

In bk.: Integral Robot Technologies and Speech Behavior. Newcastle upon Tyne: Cambridge Scholars Publishing, 2024. Ch. 9. P. 281-296.

Препринт
DAREL: Data Reduction with Losses for Training Acceleration of Real and Hypercomplex Neural Networks

Demidovskij A., Трутнев А. И., Тугарев А. М. et al.

NeurIPS 2023 Workshop. ZmuLcqwzkl. OpenReview, 2023

Reseach seminar "Applied tasks of computer vision"

2023/2024
Учебный год
ENG
Обучение ведется на английском языке
6
Кредиты

Преподаватель

Егоров Константин Сергеевич

Егоров Константин Сергеевич

Course Syllabus

Abstract

The course “Applied tasks of Computer Vision” aims to provide a basic understanding of how to tackle real-world problems. We start with the reminder on the history of computer vision and particularly convolutional neural networks. Then we dive into state-of-the-art architectures of neural networks. We elaborate on how to choose metrics and losses that are suitable for a given task. We show how to utilize widely used frameworks to set-up a pipeline in order to write scalable and reproducible code. We discuss effective and powerful post-processing tools such as uncertainty estimation, active learning, hyperparameters optimization along with classical computer vision tools. By virtue of mentioned techniques one is able to boost models’ performance as well as analyze models’ robustness. Through the course students will approach the solution of a real-world problem utilizing skills and techniques they obtain. After the completion of this course students will have insight of how to solve a problem from its formulation to the deployment of trained models.
Learning Objectives

Learning Objectives

  • Participants of this course will solve real-world computer vision problems using deep learning. In the course authors will dive into details of state-of-the-art approaches and techniques that one is able to utilize to achieve desirable performance. We will discuss theoretical as well as practical aspects of training deep neural networks. After the final exam students will have hands-on experience to solve applied tasks.
Expected Learning Outcomes

Expected Learning Outcomes

  • Be familiar with common and uncommon CV applications
  • Train simple CV model
  • Be familiar with different CV tasks and approaches
  • Be familiar with different CV metrics and losses
  • Combine CNN and GBDT models
  • Understand business metrics in a project
  • Train segmentation models
  • Write an efficient pipeline of training CV models
  • Debug models
  • Use augmentations
  • Perform postprocessing
  • Search in hyperparameters space
  • Deploy a model in production
  • Use advanced techniques to improve model’s quality and robustness
Course Contents

Course Contents

  • Introduction, history and showcases
  • CV models and losses
  • Data and evaluation
  • Optimizing training pipeline
  • Hyperparameters tuning and postprocessing
  • Ensembling, best practices, tips and tricks, production
Assessment Elements

Assessment Elements

  • non-blocking test week 1
  • non-blocking test week 2
  • non-blocking test week 3
  • non-blocking test week 4
  • non-blocking test week 5
  • non-blocking test week 6
  • non-blocking Final programming assignment
Interim Assessment

Interim Assessment

  • 2023/2024 1st module
    0.1 * test week 4 + 0.1 * test week 6 + 0.1 * test week 5 + 0.4 * Final programming assignment + 0.1 * test week 1 + 0.1 * test week 3 + 0.1 * test week 2
Bibliography

Bibliography

Recommended Core Bibliography

  • Deep learning, Goodfellow, I., 2016
  • Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). ImageNet Classification with Deep Convolutional Neural Networks. Communications of the ACM, 60(6), 84–90. https://doi.org/10.1145/3065386

Recommended Additional Bibliography

  • Изучаем OpenCV 3 : разработка программ компьютерного зрения на C++ с применением библиотеки OpenCV, Кэлер, А., 2017