Full metadata record
DC poleHodnotaJazyk
dc.contributor.authorSungatullina, Diana
dc.contributor.authorPajdla, Tomáš
dc.contributor.editorSkala, Václav
dc.date.accessioned2024-07-21T08:46:33Z-
dc.date.available2024-07-21T08:46:33Z-
dc.date.issued2024-
dc.identifier.citationJournal of WSCG. 2024, vol. 32, no. 1-2, p. 41-50.en
dc.identifier.issn1213 – 6972
dc.identifier.issn1213 – 6980 (CD-ROM)
dc.identifier.issn1213 – 6964 (on-line)
dc.identifier.urihttp://hdl.handle.net/11025/57343
dc.format10 s.cs_CZ
dc.format10 s.cs
dc.format.mimetypeapplication/pdf
dc.language.isoenen
dc.publisherVáclav Skala - UNION Agencycs
dc.rights© Václav Skala - UNION Agencycs_CZ
dc.rights© Václav Skala - UNION Agencyen
dc.subjectminimální řešitelécs
dc.subjectepipolární geometriecs
dc.subjectzpětné šířenícs
dc.subjectodstranění odlehlých hodnotcs
dc.subjectimplicitní funkcecs
dc.titleMinBackProp – Backpropagating through Minimal Solversen
dc.typečlánekcs
dc.typearticleen
dc.rights.accessopenAccessen
dc.type.versionpublishedVersion-
dc.description.abstract-translatedWe 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 slowen
dc.subject.translatedminimal solversen
dc.subject.translatedepipolar geometryen
dc.subject.translatedbackpropagationen
dc.subject.translatedoutlier removalen
dc.subject.translatedimplicit function theoremen
dc.identifier.doihttps://www.doi.org/10.24132/JWSCG.2024.5
dc.type.statusPeer-revieweden
Vyskytuje se v kolekcích:Volume 32, number 1-2 (2024)

Soubory připojené k záznamu:
Soubor Popis VelikostFormát 
A83-2024.pdfPlný text964,2 kBAdobe PDFZobrazit/otevřít
Zobrazit minimální záznam Zobrazit statistiky


Použijte tento identifikátor k citaci nebo jako odkaz na tento záznam: http://hdl.handle.net/11025/57343

Všechny záznamy v DSpace jsou chráněny autorskými právy, všechna práva vyhrazena.

hledání
navigace
  1. DSpace at University of West Bohemia
  2. Publikace ZČU / UWB Publications
  3. Časopisy ZČU / UWB Journals
  4. Journal of WSCG
  5. Volume 32 (2024)
  6. Volume 32, number 1-2 (2024)