Latest News | EIDOSLAB

This year, Santa didn’t just bring gifts, he brought us a brand new PhD 🎅📜✨

Congratulations to Gabriele Spadaro for successfully defending his PhD thesis, titled “Adaptive Compression: From Visual Data to Efficient and Transferable Models” 👏 🎓

Abstract: The exponential growth of visual content has made compression a fundamental challenge in modern communication systems. While traditional codecs achieved remarkable success, their rigid design limits their performance. Learned Image Compression emerged as a data-driven alternative, in which models directly minimize a rate-distortion loss function. Despite their results, these methods suffer from limited flexibility, since models are trained to attain a fixed rate-distortion trade-off, as well as poor generalization across novel visual domains and a lack of perceptual control. This thesis aims to investigate deep learning–based compression methods and to address the key limitations that currently hinder their deployment. Moreover, we go beyond the traditional definition of compression, proposing strategies that enhance efficiency, adaptability, and generalization capabilities by compressing the models and their internal representations.

In this context, we show how the integration of learning-based modules can significantly enhance compression performance. This improvement occurs not only by replacing specific components of standardized codecs, but also by defining end-to-end methods in which the entire compression pipeline consists of learnable modules. Interestingly, in this latter scenario, we demonstrate how the use of alternative graph-based paradigms can be effectively applied for compression tasks, while also showing their potential as general-purpose backbones for visual feature extraction.

Beyond improving compression, this thesis also proposes a unified adapter-based strategy to overcome the structural limitations of learned codecs. Considering a model-adaptation perspective, we demonstrate how adapters enable continuous control over the rate–distortion and distortion–perception trade-offs. Furthermore, they enhance the generalization capability of a pre-trained model to novel visual domains. These advances make learned codecs more versatile for heterogeneous real-world applications.

Finally, we demonstrate how model and representation pruning methods allow not only to reduce the complexity of a model, but also to improve generalization and transferability capabilities of a pre-trained model. Here, the notion of compression is extended to models and their representations. This perspective highlights its role not only as a tool for efficiency but also as a principle for designing adaptive and robust neural models.

The committee was composed by:

Enzo TARTAGLIONE – Institut Polytechnique de Paris
Thomas MAUGEY – INRIA - Université de Rennes
Federica BATTISTI – Dip. Ingegneria dell’informazione - Università degli Studi di Padova
Giulia FRACASTORO – Dip. Elettronica e Telecomunicazioni – Politecnico di Torino
Luce MORIN - Institut national des sciences appliquées de Rennes
Marco GRANGETTO – Dip. Informatica – Università di Torino

Supervisor: Attilio Fiandrotti and Enzo Tartaglione

Co-supervisors: Marco Grangetto and Jhony H. Giraldo

Well done to everyone involved and best of luck to Gabriele for the next steps of his life and research journey! 🚀

Dec 15, 2025 3 min read

Alberto Presta's PhD defense

Congratulations to Alberto Presta for successfully defending his PhD thesis, titled " Toward sensible learned image compression: closing the gap with standard codecs".

Abstract: This thesis explores limitations of current Learned Image Compression (LIC) models, particularly their inability to support multiple bitrates, progressive coding, and adaptability to unfamiliar domains—challenges not shared by traditional codecs. To address these, we introduce methods such as a parameter-free latent distribution formulation, a graph-based context estimator, and two solutions for variable rate coding: a quantization layer (STanH) and LoRA adapters for transformers. The work also proposes a dual-latent architecture for progressive compression and decoder-based domain-specific modules that enhance performance without retraining, with promising extensions to video compression.

The committee was composed by:

Giuseppe Valenzise, CNRS Researcher, CentraleSupélec Gif-sur-Yvette, France
Marco Cagnazzo, Dip. di Ingegneria dell’informazione, Uni Padova
Gabriella Olmo, Dip Informatica, Polito
Davide Cavagnino, Dip. Informatica, Unito

PhD Supervisor: Marco Grangetto (University of Turin)

May 8, 2025 1 min read

EIDOS at ECCV 2024

🌟 We had a fantastic time participating in the European Conference on Computer Vision (ECCV24), one of the top conferences in computer vision! It was a pleasure attending the event and catching up with old and new colleagues. 🤝

We’re also excited to announce that two of our papers were accepted:

📄 Boost Your NeRF: Project page Francesco Di Sario Riccardo Renzulli Enzo Tartaglione Marco Grangetto

📄 AA-SGAN: Project page Mirko Zaffaroni Federico Signoretta

Excited for what’s next! 🚀

Oct 4, 2024 1 min read

Best Italian PhD Thesis

EIDOS is very proud to announce that Andrea Bragagnolo, who got his PhD in our group, received the award for “Best Italian PhD Thesis 2023” assigned by con.Scienze. The title of his thesis is “Exploring the Design and Implementation of Pruning Techniques for Deep Neural Networks”.

Congratulations Andrea!

Jul 1, 2024 1 min read

Ayesha Hoor Chaudhry's PhD defense

Congratulations to Ayesha Hoor Chaudhry for successfully defending her PhD thesis, titled “Medical Volumetric Image Processing With Deep Neural Networks”.

The committee was composed by:

Prof. Davide CAVAGNINO – Computer Science Dept. - University of Turin;
Prof. Sinem ASLAN – Dept. of Environmental Sciences, Informatics and Statistics – Ca’ Foscari University of Venice;
Prof. Maurizio MARTINA – Dept. of Electronics and Telecommunications – Polytechnic of Turin;

PhD Supervisor: Marco Grangetto (University of Turin)

Jun 27, 2024 1 min read

EIDOS at ISBI 2024

EIDOS is very proud to contribute to ISBI 2024 with paper:

AI-Assisted Diagnosis for Covid-19 CXR Screening: From Data Collection to Clinical Validation by Carlo Alberto Barbano (University of Turin, Télécom Paris), Riccardo Renzulli (University of Turin), Marco Grosso (ASL Torino 3), Domenico Basile (ASL Torino 3), Marco Busso (ASL Torino 3), Marco Grangetto (University of Torino)

Congratulations to Carlo Barbano for presenting his work.

May 24, 2024 1 min read

EIDOS on the Air

Marco Grangetto special guest on the RBL channel EULERO show, talking about Computer Vision and prejudice.

Here is the podcast.

Apr 5, 2024 1 min read

EIDOS at DCC 2024

EIDOS is very proud to contribute to DCC 2024:

Domain Adaptation for Learned Image Compression with Supervised Adapters by Presta, Alberto; Spadaro, Gabriele; Tartaglione, Enzo; Fiandrotti, Attilio; Grangetto, Marco;

Congratulations to Alberto Presta for presenting his work with an amazing oral talk.

The Data Compression Conference (DCC) is an international forum for current work on data compression and related applications.

Mar 20, 2024 1 min read

Carlo Alberto Barbano's PhD defense

Congratulations to Carlo Alberto Barbano for successfully defending his PhD thesis, titled “Collateral-Free Learning of Deep Representations: From Natural Images to Biomedical Applications”. The committee awarded his defense with honors.

The committee was composed by:

Prof. Daniel RUECKERT – Dept. Computing - Imperial College London;
Prof. Hervé LOMBAERT - ETS Montreal;
Prof. Rosa MEO – Dip. Informatica – Università di Torino;
Prof. Enrico MAGLI – Dip. Elettronica e Telecomunicazioni – Politecnico di Torino;
Prof. Isabelle BLOCH - Télécom Paris – Institut Polytechnique de Paris;
Prof. Marco GRANGETTO - Dip. Informatica – Università di Torino;
Prof. Pietro GORI - Télécom Paris – Institut Polytechnique de Paris …

PhD Supervisor: Marco Grangetto (University of Turin), Pietro GORI (Télécom Paris – Institut Polytechnique de Paris), Isabelle BLOCH (Télécom Paris – Institut Polytechnique de Paris)

Dec 21, 2023 1 min read

Riccardo Renzulli's PhD defense

Congratulations to Riccardo Renzulli for successfully defending his PhD thesis, titled “Hierarchical Object-centric Learning with Capsule Networks”. The committee awarded his defense with honors.

The committee was composed by:

Prof. Peter Sincak, TU Kosice
Prof. Georgios Leontidis, University of Aberdeen
Prof. Lia Morra, Polytechnic University of Turin
Prof. Attilio Fiandrotti, University of Turin

PhD Supervisor: Marco Grangetto (University of Turin)

Nov 15, 2023 1 min read