EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

MGGAN : Learning a Multiscale Graph‐Based GAN for Non‐Parametric 3D Single‐Shape Deformation and Variation

Jorge Jabón, Sergio Corbera, Rafael Barea

Computer Animation and Virtual Worlds · 2026-07 · 37 (4)

DOI: 10.1002/cav.70169 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer Animation and Virtual Worlds, online erschienen am 17.07.2026. Verfasst von Jorge Jabón und 2 weiteren Autor:innen. Die Arbeit nennt 80 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Shape variation and deformation play a central role in computer animation, digital content creation and geometry processing, a domain that has undergone significant advances in the past decade owing to deep‐learning emergence. Most methods typically rely on large datasets or intermediate representations defined on Euclidean domains or grid‐like structures. As a result, learning expressive deformation models directly from surface‐based representations with limited data remains challenging. In this article, we present an unconditional learning‐based framework that captures the internal geometric statistics of a single 3D triangle mesh to generate smooth deformations ranging from local variations to larger‐scale shape changes across multiple spatial resolutions. The proposed Multiscale Graph‐based Generative Adversarial Network (MGGAN) consists of a pyramid of generators and discriminators built using spectral and spatial graph operators, trained directly on the input mesh and its downsampled versions with Gaussian noise injected at each scale. By operating directly on the input mesh, the method produces visually coherent, smooth and topology‐preserving deformed instances that are well suited for shape variation and asset exploration in animation and graphics pipelines.

Zitieren

APA

Jabón, J., Corbera, S., & Barea, R. (2026). MGGAN : Learning a Multiscale Graph‐Based GAN for Non‐Parametric 3D Single‐Shape Deformation and Variation. Computer Animation and Virtual Worlds, 37(4). https://doi.org/10.1002/cav.70169

Vancouver

Jabón J, Corbera S, Barea R. MGGAN : Learning a Multiscale Graph‐Based GAN for Non‐Parametric 3D Single‐Shape Deformation and Variation. Computer Animation and Virtual Worlds. 2026;37(4). doi:10.1002/cav.70169

BibTeX
@article{Jabn2026,
  author = {Jabón, Jorge and Corbera, Sergio and Barea, Rafael},
  title = {MGGAN : Learning a Multiscale Graph‐Based GAN for Non‐Parametric 3D Single‐Shape Deformation and Variation},
  journal = {Computer Animation and Virtual Worlds},
  year = {2026},
  volume = {37},
  number = {4},
  doi = {10.1002/cav.70169}
}
Kostenlos starten →