dc.contributor.advisor | Celledoni, Elena | |
dc.contributor.author | Riseth, Jørgen Nilsen | |
dc.date.accessioned | 2021-09-15T17:28:40Z | |
dc.date.available | 2021-09-15T17:28:40Z | |
dc.date.issued | 2021 | |
dc.identifier | no.ntnu:inspera:75366163:51739814 | |
dc.identifier.uri | https://hdl.handle.net/11250/2778386 | |
dc.description.abstract | I denne oppgaven studerer vi to gradientbaserte optimeringsalgoritmer i formanalyse, for reparametrisering av parametriske kurver og overflater. Den ene algoritmen er en tidligere kjent “gradient descent”-algoritme på gruppen av orienteringsbevarende diffeomorfier. Den andre er en ny tilnærming til parametriseringsproblemet, der det å finne en optimal reparametrisering, tilsvarer treningen av et restnevralt nettverk. Vi sammenligner ytelsen til de to algoritmene ved hjelp av noen få eksempler for både kurver og overflater. I begge tilfeller presterer det restnevrale nettverket bedre enn “gradient descent”-algoritmen. | |
dc.description.abstract | In this thesis we study two gradient-based optimization algorithms in shape analysis for reparametrization of open parametric curves and surfaces. One is a previously known Riemannian gradient descent algorithm on the group of orientation preserving diffeomorphisms. The other is a novel approach, where finding an optimal reparametrization corresponds to the training of a residual neural network. We compare the two algorithms using a few test examples for both curves and surfaces, for which the residual neural network significantly outperforms the gradient descent algorithm. | |
dc.language | eng | |
dc.publisher | NTNU | |
dc.title | Gradient-Based Optimization in Shape Analysis for Reparametrization of Parametric Curves and Surfaces | |
dc.type | Master thesis | |