Teaching code

4S010679

Teacher

Vittorio Murino

Coordinator

Vittorio Murino

Credits

6

Language

English

Scientific Disciplinary Sector (SSD)

INF/01 - INFORMATICS

Period

Semester 2 dal Mar 3, 2025 al Jun 13, 2025.

Courses Single

Authorized

Lessons timetable Moodle Seminars 0

Learning objectives

Computer vision (CV) issues are traditionally tackled by adopting machine learning methods. Recently, deep learning (DL) approaches showed to face CV applications (e.g., detection, classification, segmentation, tracking, etc.) in a most effective and efficient way, reaching performance never imagined before, even allowing to tackle new problems (e.g., image generation, style transfer, to cite a few). This course aims at describing how main CV topics are faced, and even solved, by DL approaches: It will address classical, yet significant and with broad applicability in real scenarios, CV topics, which are stil open issues, involving image and video analysis and recognition, as well as other multimodal data (3D, audio, etc.). The most significant and effective DL approaches will be detailed including, but not limited to, convolutional neural networks, autoencoder architectures, recurrent models, domain adaptation frameworks, while addressing practical problems usually met in real applications such as scarcity of annotated data (unsupervised, self-supervised, few/zero-shot learning), data augmentation and generation, robustness to adversarial attacks, and continual lifelong learning.

Examination methods

To pass the exam, students must demonstrate:
- to have understood the theoretical principles and algorithms underlying the Computer Vision & Deep Learning techniques described in class;
- to be able to present their arguments in a precise, organic and structured way, without digressions;
- to know how to apply the acquired knowledge to solve application problems presented in the form of exercises, questions or projects.
The exam will consist in the development of a project, followed by the writing of a technical report and an oral presentation.
During the oral presentation of the project, questions will be asked on the course contents described in class.

Prerequisites and basic notions

The prerequisites to follow the course consist in having acquired Machine Learning, Pattern Recognition and Artificial Intelligence skills in general.
Knowledge of topics related to image processing is also a fundamental part for understanding the course in question.

Program

The course intends to explain the modern methods for facing and solving Computational Vision problems. These methods basically consist in Deep Learning approaches applied to image and video processing.
In particular, the course will develop on monolithic topics related to specific open problems in Vision such as:
- Segmentation
- Object detection
- Object Recognition
- Image Classification
- Pose estimation
- Tracking
- Action & Activity Recognition
- Action localization
- 2D and 3D data reconstruction
- Image generation
- Image retrieval
- Multimodal data analysis
- Vision and Language
- Domain adaptation and generalization
- Transfer Learning and representation learning
- Training with scarce, noisy, unbalanced data

Didactic methods

The Theory lessons will take place in the classroom with slide projection, while the laboratory lessons will be carried out on the computer in the computer room. The latter will consist in the development of some of the algorithms explained during the Theory lectures.

Learning assessment procedures

To pass the exam, students must demonstrate:
- to have understood the theoretical principles and algorithms underlying the Computer Vision & Deep Learning techniques described in class;
- to be able to present their arguments in a precise, organic and structured way, without digressions;
- to know how to apply the acquired knowledge to solve application problems presented in the form of exercises, questions or projects.
The exam will consist in the development of a project, followed by the writing of a technical report and an oral presentation.
During the oral presentation of the project, questions will be asked on the course contents described in class.

Evaluation criteria

Quality of the project, level of difficulty and depth, quality and clarity of the project presentation.
Consistency of answers to theoretical questions.

Criteria for the composition of the final grade

A combination of the assessments related to the project and the answers to theory questions.

Exam language

Inglese