Defense of Doctoral Thesis with Soheib Takhtardeshir
Welcome to a public defense of doctoral thesis in Computer Engineering with Soheib Takhtardeshir who will present his thesis "VAE-Driven Compression of Light Field Images through Disentangled Latent Modeling and Perceptual Quality Assessment ".
About the thesis
In recent years, the demand for immersive and efficient imaging solutions has expanded with applications in fields requiring both human interpretation and automated computer vision, such as remote operations, augmented reality, and telepresence. This work addresses the challenges of representing and compressing multi-sensor video data, specifically in the context of light field imaging, which captures both angular and spatial information. Light field imaging holds potential for significantly enhancing human perception and machine vision by offering additional depth and perspective cues that traditional 2D imaging lacks. However, current light field image compression methods are often computationally intensive and unsuitable for real-time or resource-constrained applications. Our research proposes a novel variational autoencoder (VAE)-based framework for light field image compression, aimed at effectively disentangling spatial and angular information to facilitate efficient data representation and compression. This method not only achieves significant data reduction but also maintains visual fidelity, ensuring that compressed images retain essential cues for both human interpretation and algorithmic processing. By comparing our approach against state-of-the-art compression methods, we highlight its advantages in terms of processing speed and compression efficiency. Furthermore, our framework supports multi-sensor data, making it adaptable for combined human and computer vision applications. Moving forward, we will explore potential optimizations for real-time implementation and assess the framework's performance across various immersive and safety-critical applications.
More information
Date: January 22, 2026
Time: 09.15.
Place: Campus Sundsvall, Room L111, Youtube
and Zoom.
Doctoral thesis: "VAE-Driven Compression of Light Field Images through Disentangled Latent Modeling and Perceptual Quality Assessment".
Respondent: Soheib Takhtardeshir.
Supervisor and Chair: Professor Mårten Sjöström, Mid Sweden University.
Opponent/External reviewer: Professor João Ascenso, Instituto Superior Técnico, Lissabon, Portugal
Examining committee:
Docent Markus Flierl, KTH
Professor Carl Debono, University of Malta
Docent Giuseppe Valenzise, Université Paris-Saclay, Frankrike
Backup: Professor Mikael Gidlund, MId Sweden University