Umeå University's logo

umu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Deep Generative Modeling: An Overview of Recent Advances in Likelihood-based Models and an Application to 3D Point Cloud Generation
Umeå University, Faculty of Science and Technology, Department of Mathematics and Mathematical Statistics.
2023 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Deep generative modeling refers to the process of constructing a model, parameterized by a deep neural network, that learns the underlying patterns and structures of the data generating process which produced the samples in a given dataset, in order to generate novel samples that resemble those in the original dataset. Deep generative models for 3D shape generation hold significant importance to various fields including robotics, medical imaging, manufacturing, computer animation and more. This work provides a pedagogical overview of likelihood-based models, namely variational autoencoders, flow-based models, diffusion models and rectified flows and investigates the effect on generation quality of replacing the Chamfer loss with a sliced Wasserstein loss in an application to shape generation with 3D point clouds.

Abstract [sv]

Djup generativ modellering hänvisar till processen att konstruera en modell, parametriserad av ett djupt neuralt nätverk, som lär sig de underliggande mönstren och strukturerna för den datagenererande process som producerade proverna i en given datamängd, för att producera nya prover som liknar dem i den ursprungliga datamängden. Djupa generativa modeller för 3D-formgenerering har stor betydelse för olika områden, inklusive robotik, medicinsk bildbehandling, industriell tillverkning, datoranimation osv. Detta arbete ger en pedagogisk översikt av sannolikhetsbaserade modeller, nämligen variational autoencoders, flödesbaserade modeller, diffusionsmodeller och likriktade flöden och undersöker effekten på generationskvaliteten av att ersätta Chamfer-lossfunktionen med en sliced Wasserstein-lossfunktion i en applikation för att forma formgenerering med 3D-punktmoln.

Place, publisher, year, edition, pages
2023.
National Category
Mathematics
Identifiers
URN: urn:nbn:se:umu:diva-214368OAI: oai:DiVA.org:umu-214368DiVA, id: diva2:1796569
Supervisors
Examiners
Available from: 2023-09-18 Created: 2023-09-12 Last updated: 2023-09-18Bibliographically approved

Open Access in DiVA

fulltext(5235 kB)1722 downloads
File information
File name FULLTEXT01.pdfFile size 5235 kBChecksum SHA-512
657d330bcc6bbc30791b7459f1766555dbafb1c4eede0403b439676acdbab00084378d3578a64f729384a25a18961bb00f0e03da40ee12fd706f55c0a8efb7cd
Type fulltextMimetype application/pdf

By organisation
Department of Mathematics and Mathematical Statistics
Mathematics

Search outside of DiVA

GoogleGoogle Scholar
Total: 1723 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 713 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf