MinBackProp – Backpropagating through Minimal Solvers

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F24%3A00377151" target="_blank" >RIV/68407700:21230/24:00377151 - isvavai.cz</a>

Alternative codes found

RIV/68407700:21730/24:00377151

Result on the web

<a href="https://doi.org/10.24132/JWSCG.2024.5" target="_blank" >https://doi.org/10.24132/JWSCG.2024.5</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.24132/JWSCG.2024.5" target="_blank" >10.24132/JWSCG.2024.5</a>

Result language

angličtina

Original language name

MinBackProp – Backpropagating through Minimal Solvers

Original language description

We present an approach to backpropagating through minimal problem solvers in end-to-end neural network train ing. Traditional methods relying on manually constructed formulas, finite differences, and autograd are laborious, approximate, and unstable for complex minimal problem solvers. We show that using the Implicit function the orem (IFT) to calculate derivatives to backpropagate through the solution of a minimal problem solver is simple, fast, and stable. We compare our approach to (i) using the standard autograd on minimal problem solvers and relate it to existing backpropagation formulas through SVD-based and Eig-based solvers and (ii) implementing the backprop with an existing PyTorch Deep Declarative Networks (DDN) framework [GHC22]. We demonstrate our technique on a toy example of training outlier-rejection weights for 3D point registration and on a real application of training an outlier-rejection and RANSAC sampling network in image matching. Our method provides 100% stability and is 10 times faster compared to autograd, which is unstable and slow, and compared to DDN, which is stable but also slow.

Czech name

—

Czech description

—

Type

J_SC - Article in a specialist periodical, which is included in the SCOPUS database

CEP classification

—

OECD FORD branch

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

Project

—

Continuities

R - Projekt Ramcoveho programu EK

Publication year

2024

Confidentiality

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

Name of the periodical

Journal of WSCG

ISSN

1213-6972

e-ISSN

—

Volume of the periodical

32

Issue of the periodical within the volume

1-2

Country of publishing house

CZ - CZECH REPUBLIC

Number of pages

10

Pages from-to

41-50

UT code for WoS article

—

EID of the result in the Scopus database

2-s2.0-85203719859