C9749 - DEEP LEARNING TECHNIQUES FOR COMPUTATIONAL IMAGING

Learning outcomes

At the end of the course the student knows about computational imaging methods and applications with a focus on solving inverse problems in imaging, such as denoising, deconvolution, single-pixel imaging, and others. He can solve some of the previous imaging problems by using both classic optimization algorithms and modern data-driven approaches with convolutional neural networks (CNNs).

Course contents

Basic concepts of image formation and noise modeling

– Mathematical tools for image processing: filters, discrete Fourier transform

– Computational imaging applications as inverse problems: denoising, deblurring, super-resolution, segmentation, tomographic reconstruction, …

– Classical methods based on regularization for solving computational imaging problems

– Convolutional neural network approaches: study of architectures and state-of-the-art imaging losses

– Generative approaches: Generative Adversarial Networks (GANs), Diffusion Models, and their applications in computational imaging. Recent developments

– Hands-on sessions using Python and PyTorch

Readings/Bibliography

notes of the teachers

Teaching methods

Lectures and guided labs

Assessment methods

The examination consists of two parts:

• a written examination, consisting of one open-ended question on the course topics (no programming code), lasting 20 minutes. The maximum score is 8 points, and the minimum passing score is 4 points;

• an oral discussion of the group project, with a maximum score of 24 points and a minimum passing score of 14 points.

The final grade is the sum of the scores obtained in the two parts. If the total score exceeds 30, the student is awarded 30 cum laude.

All members of the same project group are required to take the examination on the same day.

The following rules apply:

1. If the project discussion is passed but the written examination is failed, the student must retake only the written examination.

2. If the project discussion is failed, the student must retake the entire examination.

For the project discussion, students must prepare a presentation including the theoretical background, implementation details, and the results obtained, presented in the form of figures and tables. The presentation must not include the source code. However, students are required to bring the project code to the examination, as they will be asked questions about it.

The use of AI tools for code development is permitted. However, during the project discussion, students must demonstrate a thorough understanding of their code, be able to explain how it works, and justify the implementation choices they made.

Office hours

See the website of Elena Loli Piccolomini

See the website of Davide Evangelista