SUPR
Representation and Compression of Multi-sensor Video for Combined Human and Computer Vision Applications
Dnr:

NAISS 2024/22-126

Type:

NAISS Small Compute

Principal Investigator:

Soheib Takhtardeshir

Affiliation:

Mittuniversitetet

Start Date:

2024-01-24

End Date:

2025-02-01

Primary Classification:

10206: Computer Engineering

Allocation

Abstract

Inserting deep learning tools into the framework of the versatile video coding (VVC) engineered approach to provide enhanced RD performance from signal-dependent compression without increasing total computational complexity. Develop a compression framework based on variational autoencoder and generative adversarial networks (GAN) architectures decoupled from the traditional hybrid coding paradigm and fully adopt signal-dependent compression. Incorporate scene flow to enhance prediction accuracy for any temporal redundancy reduction system within the above two approaches. For the mentioned aims, I need to work with DNN networks, and it requires High-performance Computing systems.