Accurate Optimization of Weighted Nuclear Norm for Non-Rigid Structure from Motion
Paper i proceeding, 2020

Fitting a matrix of a given rank to data in a least squares sense can be done very effectively using 2nd order methods such as Levenberg-Marquardt by explicitly optimizing over a bilinear parameterization of the matrix. In contrast, when applying more general singular value penalties, such as weighted nuclear norm priors, direct optimization over the elements of the matrix is typically used. Due to non-differentiability of the resulting objective function, first order sub-gradient or splitting methods are predominantly used. While these offer rapid iterations it is well known that they become inefficent near the minimum due to zig-zagging and in practice one is therefore often forced to settle for an approximate solution. In this paper we show that more accurate results can in many cases be achieved with 2nd order methods. Our main result shows how to construct bilinear formulations, for a general class of regularizers including weighted nuclear norm penalties, that are provably equivalent to the original problems. With these formulations the regularizing function becomes twice differentiable and 2nd order methods can be applied. We show experimentally, on a number of structure from motion problems, that our approach outperforms state-of-the-art methods.

Författare

José Pedro Lopes Iglesias

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

Carl Olsson

Lunds universitet

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

Marcus Valtonen Örnhag

Lunds universitet

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

03029743 (ISSN) 16113349 (eISSN)

Vol. 12372 LNCS 21-37
9783030585822 (ISBN)

European Conference on Computer Vision
, ,

Optimeringsmetoder med prestandagarantier för maskininlärningsmetoder

Vetenskapsrådet (VR) (2018-05375), 2019-01-01 -- 2022-12-31.

Ämneskategorier

Beräkningsmatematik

Reglerteknik

Matematisk analys

DOI

10.1007/978-3-030-58583-9_2

ISBN

9783030585822

Mer information

Senast uppdaterat

2024-09-23