Neural weight step video compression

Mikolaj Czerkawski; Javier Cardona; Robert Atkinson; Craig Michie; Ivan Andonovic; Carmine Clemente; Christos Tachtatzis

Neural weight step video compression

Mikolaj Czerkawski, Javier Cardona, Robert Atkinson, Craig Michie, Ivan Andonovic, Carmine Clemente, Christos Tachtatzis

Research output: Contribution to conference › Paper › peer-review

39 Downloads (Pure)

Abstract

A variety of compression methods based on encoding images as weights of a neural network have been recently proposed. Yet, the potential of similar approaches for video compression remains unexplored. In this work, we suggest a set of experiments for testing the feasibility of compressing video using two architectural paradigms, coordinate-based MLP (CbMLP) and convolutional network. Furthermore, we propose a novel technique of neural weight stepping, where subsequent frames of a video are encoded as low-entropy parameter updates. To assess the feasibility of the considered approaches, we will test the video compression performance on several high-resolution video datasets and compare against existing conventional and neural compression techniques.

Original language	English
Number of pages	8
Publication status	Published - 13 Dec 2021
Event	NeurIPS 2021 Pre-Registration Workshop: An Alternative Publication Model for Machine Learning research - Online Duration: 13 Dec 2021 → 13 Dec 2021 https://preregister.science

Workshop

Workshop	NeurIPS 2021 Pre-Registration Workshop
Abbreviated title	PreReg@NeurIPS'21
Period	13/12/21 → 13/12/21
Internet address	https://preregister.science

Keywords

image encoding
machine learning
neural networks
CbMLP
video compression

Access to Document

Czerkawski-etal-PreReg-2021-Neural-weight-step-video-compressionFinal published version, 168 KBLicence: CC BY-NC-ND 4.0
Czerkawski-etal-PreReg-2021-Neural-weight-step-video-compression-video-summaryOther version, 7.15 MBLicence: CC BY-NC-ND 4.0

Doctoral Training Partnership 2018-19 University of Strathclyde | Czerkawski, Mikolaj
Tachtatzis, C. (Principal Investigator), Clemente, C. (Co-investigator) & Czerkawski, M. (Research Co-investigator)
EPSRC (Engineering and Physical Sciences Research Council)
1/10/19 → 11/08/23
Project: Research Studentship - Internally Allocated

Cite this

@conference{1b8eca859e584b4c85fb7491b7dfaf47,

title = "Neural weight step video compression",

abstract = "A variety of compression methods based on encoding images as weights of a neural network have been recently proposed. Yet, the potential of similar approaches for video compression remains unexplored. In this work, we suggest a set of experiments for testing the feasibility of compressing video using two architectural paradigms, coordinate-based MLP (CbMLP) and convolutional network. Furthermore, we propose a novel technique of neural weight stepping, where subsequent frames of a video are encoded as low-entropy parameter updates. To assess the feasibility of the considered approaches, we will test the video compression performance on several high-resolution video datasets and compare against existing conventional and neural compression techniques.",

keywords = "image encoding, machine learning, neural networks, CbMLP, video compression",

author = "Mikolaj Czerkawski and Javier Cardona and Robert Atkinson and Craig Michie and Ivan Andonovic and Carmine Clemente and Christos Tachtatzis",

year = "2021",

month = dec,

day = "13",

language = "English",

note = "NeurIPS 2021 Pre-Registration Workshop : An Alternative Publication Model for Machine Learning research, PreReg@NeurIPS'21 ; Conference date: 13-12-2021 Through 13-12-2021",

url = "https://preregister.science",

}

TY - CONF

T1 - Neural weight step video compression

AU - Czerkawski, Mikolaj

AU - Cardona, Javier

AU - Atkinson, Robert

AU - Michie, Craig

AU - Andonovic, Ivan

AU - Clemente, Carmine

AU - Tachtatzis, Christos

PY - 2021/12/13

Y1 - 2021/12/13

N2 - A variety of compression methods based on encoding images as weights of a neural network have been recently proposed. Yet, the potential of similar approaches for video compression remains unexplored. In this work, we suggest a set of experiments for testing the feasibility of compressing video using two architectural paradigms, coordinate-based MLP (CbMLP) and convolutional network. Furthermore, we propose a novel technique of neural weight stepping, where subsequent frames of a video are encoded as low-entropy parameter updates. To assess the feasibility of the considered approaches, we will test the video compression performance on several high-resolution video datasets and compare against existing conventional and neural compression techniques.

AB - A variety of compression methods based on encoding images as weights of a neural network have been recently proposed. Yet, the potential of similar approaches for video compression remains unexplored. In this work, we suggest a set of experiments for testing the feasibility of compressing video using two architectural paradigms, coordinate-based MLP (CbMLP) and convolutional network. Furthermore, we propose a novel technique of neural weight stepping, where subsequent frames of a video are encoded as low-entropy parameter updates. To assess the feasibility of the considered approaches, we will test the video compression performance on several high-resolution video datasets and compare against existing conventional and neural compression techniques.

KW - image encoding

KW - machine learning

KW - neural networks

KW - CbMLP

KW - video compression

M3 - Paper

T2 - NeurIPS 2021 Pre-Registration Workshop

Y2 - 13 December 2021 through 13 December 2021

ER -

Neural weight step video compression

Abstract

Workshop

Keywords

Access to Document

Fingerprint

Projects

Doctoral Training Partnership 2018-19 University of Strathclyde | Czerkawski, Mikolaj

Cite this