Název:	MinBackProp – Backpropagating through Minimal Solvers
Autoři:	Sungatullina, Diana Pajdla, Tomáš
Citace zdrojového dokumentu:	Journal of WSCG. 2024, vol. 32, no. 1-2, p. 41-50.
Datum vydání:	2024
Nakladatel:	Václav Skala - UNION Agency
Typ dokumentu:	článek article
URI:	http://hdl.handle.net/11025/57343
ISSN:	1213 – 6972 1213 – 6980 (CD-ROM) 1213 – 6964 (on-line)
Klíčová slova:	minimální řešitelé;epipolární geometrie;zpětné šíření;odstranění odlehlých hodnot;implicitní funkce
Klíčová slova v dalším jazyce:	minimal solvers;epipolar geometry;backpropagation;outlier removal;implicit function theorem
Abstrakt v dalším jazyce:	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
Práva:	© Václav Skala - UNION Agency © Václav Skala - UNION Agency
Vyskytuje se v kolekcích:	Volume 32, number 1-2 (2024)

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