Light-Weight EPINET Architecture for Fast Light Field Disparity Estimation

Ali Hassan; Marten Sjostrom; Tingting Zhang; Karen Egiazarian

doi:10.1109/MMSP55362.2022.9949378

Light-Weight EPINET Architecture for Fast Light Field Disparity Estimation

Ali Hassan, Marten Sjostrom, Tingting Zhang, Karen Egiazarian

Tietotekniikka

Tutkimustuotos: Konferenssiartikkeli › Tieteellinen › vertaisarvioitu

5 Sitaatiot (Scopus)

34 Lataukset (Pure)

Abstrakti

Recent deep learning-based light field disparity estimation algorithms require millions of parameters, which demand high computational cost and limit the model deployment. In this paper, an investigation is carried out to analyze the effect of depthwise separable convolution and ghost modules on state-of-the-art EPINET architecture for disparity estimation. Based on this investigation, four convolutional blocks are proposed to make the EPINET architecture a fast and light-weight network for disparity estimation. The experimental results exhibit that the proposed convolutional blocks have significantly reduced the computational cost of EPINET architecture by up to a factor of 3.89, while achieving comparable disparity maps on HCI Benchmark dataset.

Alkuperäiskieli	Englanti
Otsikko	2022 IEEE 24th International Workshop on Multimedia Signal Processing, MMSP 2022
Kustantaja	IEEE
ISBN (elektroninen)	9781665471893
DOI - pysyväislinkit	https://doi.org/10.1109/MMSP55362.2022.9949378
Tila	Julkaistu - 2022
OKM-julkaisutyyppi	A4 Artikkeli konferenssijulkaisussa
Tapahtuma	IEEE International Workshop on Multimedia Signal Processing - Shanghai, Kiina Kesto: 26 syysk. 2022 → 28 syysk. 2022

Julkaisusarja

Nimi	IEEE International Workshop on Multimedia Signal Processing, MMSP
ISSN (elektroninen)	2473-3628

Conference

Conference	IEEE International Workshop on Multimedia Signal Processing
Maa/Alue	Kiina
Kaupunki	Shanghai
Ajanjakso	26/09/22 → 28/09/22

Julkaisufoorumi-taso

Jufo-taso 1

!!ASJC Scopus subject areas

Computer Vision and Pattern Recognition
Signal Processing
Media Technology

Pääsy asiakirjaan

10.1109/MMSP55362.2022.9949378

Light-Weight EPINET Architecture
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Accepted author manuscript, 1,17 MBLisenssi: Unspecified

https://urn.fi/URN:NBN:fi:tuni-202212229735Lisenssi: Unspecified

Siteeraa tätä

@inproceedings{8e3e1216032b44e7b31389510012ca88,

title = "Light-Weight EPINET Architecture for Fast Light Field Disparity Estimation",

abstract = "Recent deep learning-based light field disparity estimation algorithms require millions of parameters, which demand high computational cost and limit the model deployment. In this paper, an investigation is carried out to analyze the effect of depthwise separable convolution and ghost modules on state-of-the-art EPINET architecture for disparity estimation. Based on this investigation, four convolutional blocks are proposed to make the EPINET architecture a fast and light-weight network for disparity estimation. The experimental results exhibit that the proposed convolutional blocks have significantly reduced the computational cost of EPINET architecture by up to a factor of 3.89, while achieving comparable disparity maps on HCI Benchmark dataset.",

keywords = "Compression, Deep Learning, Depthwise Separable Convolution, Disparity Estimation, Light Field",

author = "Ali Hassan and Marten Sjostrom and Tingting Zhang and Karen Egiazarian",

note = "Funding Information: The work was supported by the European Joint Doctoral Programme on Plenoptic Imaging (PLENOPTIMA) through the European Union s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant Agreement No. 956770 Funding Information: The work was supported by the European Joint Doctoral Programme on Plenoptic Imaging (PLENOPTIMA) through the European Union{\textquoteright}s Horizon 2020 research and innovation programme under the Marie Sk{\l}odowska-Curie Grant Agreement No. 956770. Publisher Copyright: {\textcopyright} 2022 IEEE. jufoid=70574; IEEE International Workshop on Multimedia Signal Processing ; Conference date: 26-09-2022 Through 28-09-2022",

year = "2022",

doi = "10.1109/MMSP55362.2022.9949378",

language = "English",

series = "IEEE International Workshop on Multimedia Signal Processing, MMSP ",

publisher = "IEEE",

booktitle = "2022 IEEE 24th International Workshop on Multimedia Signal Processing, MMSP 2022",

address = "United States",

}

Hassan, A, Sjostrom, M, Zhang, T & Egiazarian, K 2022, Light-Weight EPINET Architecture for Fast Light Field Disparity Estimation. julkaisussa 2022 IEEE 24th International Workshop on Multimedia Signal Processing, MMSP 2022. IEEE International Workshop on Multimedia Signal Processing, MMSP , IEEE, IEEE International Workshop on Multimedia Signal Processing, Shanghai, Kiina, 26/09/22. https://doi.org/10.1109/MMSP55362.2022.9949378

TY - GEN

T1 - Light-Weight EPINET Architecture for Fast Light Field Disparity Estimation

AU - Hassan, Ali

AU - Sjostrom, Marten

AU - Zhang, Tingting

AU - Egiazarian, Karen

N1 - Funding Information: The work was supported by the European Joint Doctoral Programme on Plenoptic Imaging (PLENOPTIMA) through the European Union s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant Agreement No. 956770 Funding Information: The work was supported by the European Joint Doctoral Programme on Plenoptic Imaging (PLENOPTIMA) through the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 956770. Publisher Copyright: © 2022 IEEE. jufoid=70574

PY - 2022

Y1 - 2022

N2 - Recent deep learning-based light field disparity estimation algorithms require millions of parameters, which demand high computational cost and limit the model deployment. In this paper, an investigation is carried out to analyze the effect of depthwise separable convolution and ghost modules on state-of-the-art EPINET architecture for disparity estimation. Based on this investigation, four convolutional blocks are proposed to make the EPINET architecture a fast and light-weight network for disparity estimation. The experimental results exhibit that the proposed convolutional blocks have significantly reduced the computational cost of EPINET architecture by up to a factor of 3.89, while achieving comparable disparity maps on HCI Benchmark dataset.

AB - Recent deep learning-based light field disparity estimation algorithms require millions of parameters, which demand high computational cost and limit the model deployment. In this paper, an investigation is carried out to analyze the effect of depthwise separable convolution and ghost modules on state-of-the-art EPINET architecture for disparity estimation. Based on this investigation, four convolutional blocks are proposed to make the EPINET architecture a fast and light-weight network for disparity estimation. The experimental results exhibit that the proposed convolutional blocks have significantly reduced the computational cost of EPINET architecture by up to a factor of 3.89, while achieving comparable disparity maps on HCI Benchmark dataset.

KW - Compression

KW - Deep Learning

KW - Depthwise Separable Convolution

KW - Disparity Estimation

KW - Light Field

U2 - 10.1109/MMSP55362.2022.9949378

DO - 10.1109/MMSP55362.2022.9949378

M3 - Conference contribution

AN - SCOPUS:85143584260

T3 - IEEE International Workshop on Multimedia Signal Processing, MMSP

BT - 2022 IEEE 24th International Workshop on Multimedia Signal Processing, MMSP 2022

PB - IEEE

T2 - IEEE International Workshop on Multimedia Signal Processing

Y2 - 26 September 2022 through 28 September 2022

ER -

Light-Weight EPINET Architecture for Fast Light Field Disparity Estimation

Abstrakti

Julkaisusarja

Conference

Julkaisufoorumi-taso

!!ASJC Scopus subject areas

Pääsy asiakirjaan

Sormenjälki

Siteeraa tätä