Halvtidsseminarium med Ali Hassan

Title: Dimension Reduction of Deep Network Models for Mobile Computational Plenoptic

Respondent: Ali Hassan

Opponent: Assoc. Prof. Alexandros Sopasakis (Lund University, Sweden)

Supervisors: Prof. Mårten Sjöström (Mid Sweden University), Prof. Karen Egiazarian (Tampere University, Finland) and Prof. Tingting Zhang (Mid Sweden University)

Places: Campus Sundsvall, L408 or via Teams

Time: 09:00-11:00

The half time seminar will be held in English. 

Abstract

Recent advancements in image acquisition technology have facilitated the capture of Light Field (LF) data, which encompasses both the spatial and angular information of a scene. This capability allows users to manipulate scene information, enhancing immersive experience in multimedia applications. Majority of these manipulations and post-processing tasks, such as LF disparity estimation, view synthesis and super-resolution, are increasingly performed using deep learning techniques. However, the state-of-the-art deep learning architecture often requires millions of parameters, making them energy, computation and memory intensive. Consequently, these computational demands limit their applicability on mobile devices. This report explores two different perspectives to reduce the complexity of deep learning architectures while optimizing their performance. By reducing the time and effort required to find the optimal architecture, this research opens up new opportunities for the community, making advanced computer vision more accessible in complex applications, such as multi-dimensional light field technology.