Mixed Precision ℓ1 Solver for Compressive Depth Reconstruction: An ADMM Case Study

Yun Wu; Andrew Michael Wallace; Andreas Aßmann; Brian Stewart

doi:10.1109/SiPS52927.2021.00021

Mixed Precision ℓ1 Solver for Compressive Depth Reconstruction: An ADMM Case Study

Yun Wu, Andrew Michael Wallace, Andreas Aßmann, Brian Stewart

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Rapid reconstruction of depth images from sparsely sampled data is important for many machine learning applications, including robot or vehicle assistance or autonomy, which require low power LiDAR sensing for eye safety, and resource reduction for FPGA or solid state implementation, especially with constrained energy budgets. A new compressive depth reconstruction design approach is proposed using a compact ADMM solver for the lasso problem, which varies the precision scaling in an iterative optimization process. Implementations on an FPGA architecture show over 55% savings in hardware resources and 78% in power with only minor reduction in reconstructed depth image quality compared to single float precision.

Original language	English
Title of host publication	2021 IEEE Workshop on Signal Processing Systems (SiPS)
Publisher	IEEE
Pages	70-75
Number of pages	6
ISBN (Electronic)	9781665401449
DOIs	https://doi.org/10.1109/SiPS52927.2021.00021
Publication status	Published - 13 Nov 2021
Event	2021 International Workshop on Signal Processing Systems - Coimbra, Portugal Duration: 19 Oct 2021 → 21 Oct 2021 https://sips2021.org/

Workshop

Workshop	2021 International Workshop on Signal Processing Systems
Abbreviated title	SiPS 2021
Country/Territory	Portugal
City	Coimbra
Period	19/10/21 → 21/10/21
Internet address	https://sips2021.org/

Keywords

Alternating Direction Method of Multipliers
Compressive Sensing
Depth Reconstruction
Field-Programmable Gate Array
Mixed Precision

ASJC Scopus subject areas

Electrical and Electronic Engineering
Signal Processing
Applied Mathematics
Hardware and Architecture

Access to Document

10.1109/SiPS52927.2021.00021

Cite this

@inproceedings{d80291ee41ba4f55b7f8ab9829042d89,

title = "Mixed Precision ℓ1 Solver for Compressive Depth Reconstruction: An ADMM Case Study",

abstract = "Rapid reconstruction of depth images from sparsely sampled data is important for many machine learning applications, including robot or vehicle assistance or autonomy, which require low power LiDAR sensing for eye safety, and resource reduction for FPGA or solid state implementation, especially with constrained energy budgets. A new compressive depth reconstruction design approach is proposed using a compact ADMM solver for the lasso problem, which varies the precision scaling in an iterative optimization process. Implementations on an FPGA architecture show over 55% savings in hardware resources and 78% in power with only minor reduction in reconstructed depth image quality compared to single float precision. ",

keywords = "Alternating Direction Method of Multipliers, Compressive Sensing, Depth Reconstruction, Field-Programmable Gate Array, Mixed Precision",

author = "Yun Wu and Wallace, {Andrew Michael} and Andreas A{\ss}mann and Brian Stewart",

note = "Funding Information: VI. ACKNOWLEDGEMENT This work is supported by the Engineering and Physical Sciences Research Council of the UK (EPSRC) Grant number EP/S000631/1 and the UK MOD University Defence Research Collaboration (UDRC) in Signal Processing Publisher Copyright: {\textcopyright} 2021 IEEE.; 2021 International Workshop on Signal Processing Systems, SiPS 2021 ; Conference date: 19-10-2021 Through 21-10-2021",

year = "2021",

month = nov,

day = "13",

doi = "10.1109/SiPS52927.2021.00021",

language = "English",

pages = "70--75",

booktitle = "2021 IEEE Workshop on Signal Processing Systems (SiPS)",

publisher = "IEEE",

address = "United States",

url = "https://sips2021.org/",

}

TY - GEN

T1 - Mixed Precision ℓ1 Solver for Compressive Depth Reconstruction

T2 - 2021 International Workshop on Signal Processing Systems

AU - Wu, Yun

AU - Wallace, Andrew Michael

AU - Aßmann, Andreas

AU - Stewart, Brian

N1 - Funding Information: VI. ACKNOWLEDGEMENT This work is supported by the Engineering and Physical Sciences Research Council of the UK (EPSRC) Grant number EP/S000631/1 and the UK MOD University Defence Research Collaboration (UDRC) in Signal Processing Publisher Copyright: © 2021 IEEE.

PY - 2021/11/13

Y1 - 2021/11/13

N2 - Rapid reconstruction of depth images from sparsely sampled data is important for many machine learning applications, including robot or vehicle assistance or autonomy, which require low power LiDAR sensing for eye safety, and resource reduction for FPGA or solid state implementation, especially with constrained energy budgets. A new compressive depth reconstruction design approach is proposed using a compact ADMM solver for the lasso problem, which varies the precision scaling in an iterative optimization process. Implementations on an FPGA architecture show over 55% savings in hardware resources and 78% in power with only minor reduction in reconstructed depth image quality compared to single float precision.

AB - Rapid reconstruction of depth images from sparsely sampled data is important for many machine learning applications, including robot or vehicle assistance or autonomy, which require low power LiDAR sensing for eye safety, and resource reduction for FPGA or solid state implementation, especially with constrained energy budgets. A new compressive depth reconstruction design approach is proposed using a compact ADMM solver for the lasso problem, which varies the precision scaling in an iterative optimization process. Implementations on an FPGA architecture show over 55% savings in hardware resources and 78% in power with only minor reduction in reconstructed depth image quality compared to single float precision.

KW - Alternating Direction Method of Multipliers

KW - Compressive Sensing

KW - Depth Reconstruction

KW - Field-Programmable Gate Array

KW - Mixed Precision

UR - http://www.scopus.com/inward/record.url?scp=85122888181&partnerID=8YFLogxK

U2 - 10.1109/SiPS52927.2021.00021

DO - 10.1109/SiPS52927.2021.00021

M3 - Conference contribution

AN - SCOPUS:85122888181

SP - 70

EP - 75

BT - 2021 IEEE Workshop on Signal Processing Systems (SiPS)

PB - IEEE

Y2 - 19 October 2021 through 21 October 2021

ER -

Mixed Precision ℓ1 Solver for Compressive Depth Reconstruction: An ADMM Case Study

Abstract

Workshop

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this