MPLP++: Fast, Parallel Dual Block-Coordinate Ascent for Dense Graphical Models

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F18%3A00326227" target="_blank" >RIV/68407700:21230/18:00326227 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1007/978-3-030-01225-0_16" target="_blank" >http://dx.doi.org/10.1007/978-3-030-01225-0_16</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-030-01225-0_16" target="_blank" >10.1007/978-3-030-01225-0_16</a>

Jazyk výsledku

angličtina

Název v původním jazyce

MPLP++: Fast, Parallel Dual Block-Coordinate Ascent for Dense Graphical Models

Popis výsledku v původním jazyce

Dense, discrete Graphical Models with pairwise potentials are a powerful class of models which are employed in state-of-the-art computer vision and bio-imaging applications. This work introduces a new MAP-solver, based on the popular Dual Block-Coordinate Ascent principle. Surprisingly, by making a small change to a low-performing solver, the Max Product Linear Programming (MPLP) algorithm [7], we derive the new solver MPLP++ that significantly outperforms all existing solvers by a large margin, including the state-of-the-art solver Tree-Reweighted Sequential (TRW-S) message-passing algorithm [17]. Additionally, our solver is highly parallel, in contrast to TRW-S, which gives a further boost in performance with the proposed GPU and multi-thread CPU implementations. We verify the superiority of our algorithm on dense problems from publicly available benchmarks as well as a new benchmark for 6D Object Pose estimation. We also provide an ablation study with respect to graph density.

Název v anglickém jazyce

MPLP++: Fast, Parallel Dual Block-Coordinate Ascent for Dense Graphical Models

Popis výsledku anglicky

Dense, discrete Graphical Models with pairwise potentials are a powerful class of models which are employed in state-of-the-art computer vision and bio-imaging applications. This work introduces a new MAP-solver, based on the popular Dual Block-Coordinate Ascent principle. Surprisingly, by making a small change to a low-performing solver, the Max Product Linear Programming (MPLP) algorithm [7], we derive the new solver MPLP++ that significantly outperforms all existing solvers by a large margin, including the state-of-the-art solver Tree-Reweighted Sequential (TRW-S) message-passing algorithm [17]. Additionally, our solver is highly parallel, in contrast to TRW-S, which gives a further boost in performance with the proposed GPU and multi-thread CPU implementations. We verify the superiority of our algorithm on dense problems from publicly available benchmarks as well as a new benchmark for 6D Object Pose estimation. We also provide an ablation study with respect to graph density.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/GA18-25383S" target="_blank" >GA18-25383S: Učící modely pro úlohu hledání hustých korespondencí v obrazech</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2018

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

ECCV2018: Proceedings of the European Conference on Computer Vision, Part IV

ISBN

978-3-030-01224-3

ISSN

—

e-ISSN

0302-9743

Počet stran výsledku

18

Strana od-do

264-281

Název nakladatele

Springer-Verlag

Místo vydání

Berlin

Místo konání akce

Munich

Datum konání akce

8. 9. 2018

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—