Full metadata record
DC pole | Hodnota | Jazyk |
---|---|---|
dc.contributor.author | Sungatullina, Diana | |
dc.contributor.author | Pajdla, Tomáš | |
dc.contributor.editor | Skala, Václav | |
dc.date.accessioned | 2024-07-21T08:46:33Z | - |
dc.date.available | 2024-07-21T08:46:33Z | - |
dc.date.issued | 2024 | - |
dc.identifier.citation | Journal of WSCG. 2024, vol. 32, no. 1-2, p. 41-50. | en |
dc.identifier.issn | 1213 – 6972 | |
dc.identifier.issn | 1213 – 6980 (CD-ROM) | |
dc.identifier.issn | 1213 – 6964 (on-line) | |
dc.identifier.uri | http://hdl.handle.net/11025/57343 | |
dc.format | 10 s. | cs_CZ |
dc.format | 10 s. | cs |
dc.format.mimetype | application/pdf | |
dc.language.iso | en | en |
dc.publisher | Václav Skala - UNION Agency | cs |
dc.rights | © Václav Skala - UNION Agency | cs_CZ |
dc.rights | © Václav Skala - UNION Agency | en |
dc.subject | minimální řešitelé | cs |
dc.subject | epipolární geometrie | cs |
dc.subject | zpětné šíření | cs |
dc.subject | odstranění odlehlých hodnot | cs |
dc.subject | implicitní funkce | cs |
dc.title | MinBackProp – Backpropagating through Minimal Solvers | en |
dc.type | článek | cs |
dc.type | article | en |
dc.rights.access | openAccess | en |
dc.type.version | publishedVersion | - |
dc.description.abstract-translated | 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 | en |
dc.subject.translated | minimal solvers | en |
dc.subject.translated | epipolar geometry | en |
dc.subject.translated | backpropagation | en |
dc.subject.translated | outlier removal | en |
dc.subject.translated | implicit function theorem | en |
dc.identifier.doi | https://www.doi.org/10.24132/JWSCG.2024.5 | |
dc.type.status | Peer-reviewed | en |
Vyskytuje se v kolekcích: | Volume 32, number 1-2 (2024) |
Soubory připojené k záznamu:
Soubor | Popis | Velikost | Formát | |
---|---|---|---|---|
A83-2024.pdf | Plný text | 964,2 kB | Adobe PDF | Zobrazit/otevřít |
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.