Differentiable 3D Scene Representations With Point-Based Neural Methods
Guardado en:
| Publicado en: | ProQuest Dissertations and Theses (2025) |
|---|---|
| Autor principal: | |
| Publicado: |
ProQuest Dissertations & Theses
|
| Materias: | |
| Acceso en línea: | Citation/Abstract Full Text - PDF |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Resumen: | This thesis explores reconstructing explicit scene geometry via geometry particles carrying local Lagrangian patches. We formulate a signed distance field as a weighted sum of moving basis functions and describe an optimization framework to fit target shapes in both 2D and 3D. Experiments on canonical geometry meshes show that with a modest number of particles, our approach can capture coarse geometric structures while providing intuitive control and interpretable local geometry images in a storage-efficient representation. Although these preliminary results do not yet match state-of-the-art accuracy, they highlight the promise of a particle-based, differentiable explicit representation that is suitable to inspire further work in a vast array of workflow improvements from digital sculpting to generative modeling. We conclude by discussing avenues for further research on improving particle placement, blending strategies, and interactive editing capabilities. |
|---|---|
| ISBN: | 9798263351267 |
| Fuente: | ProQuest Dissertations & Theses Global |