Research map / 3D vision / Face and shape
Gaussian Process Morphable Models
Marcel Lüthi, Thomas Gerig, Christoph Jud, Thomas R. Vetter
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use anisotropic multiscale kernel instead of single Gaussian kernel (ablation) | GPMM ulna model · Ulna shape model (full model) | specificity: 5.8 -> 2.9 | Improved |
| use anisotropic multiscale kernel instead of isotropic multiscale kernel (ablation) | GPMM ulna model · Ulna shape model (full model) | compactness: 317.0 -> 137.1 | Improved |
| use anisotropic multiscale kernel instead of single Gaussian kernel (ablation) | GPMM ulna model · Ulna shape model (full model) | compactness: 299.1 -> 137.1 | Improved |
| use isotropic multiscale kernel instead of single Gaussian kernel (ablation) | GPMM ulna model · Ulna shape model (full model) | specificity: 5.8 -> 6.1 | Hurt |
| use multiscale or anisotropic kernel instead of single Gaussian kernel (ablation) | GPMM · Ulna surface registration | generalization | Improved |
| add smooth additive deformation or localized kernel to ASM shape model (ablation) | Active Shape Model · Ulna ASM fitting (leave-one-out) | fitting accuracy | Mixed |
| add 4 landmark constraints via posterior model to extended ASM (ablation) | GPMM Active Shape Model · Ulna ASM fitting (leave-one-out) | fitting accuracy | Improved |
| use anisotropic model instead of Gaussian or multiscale in image registration (ablation) | GPMM · Ulna image registration | surface distance | Improved |
| use multi-resolution B-spline instead of GPMM | image registration · Ulna image registration | accuracy: 500 parameters vs 37926 parameters | Hurt |
| use single-level B-spline instead of GPMM | image registration · Ulna image registration | accuracy and robustness | Hurt |
| increase Nystrom points from n=200 to n=1000 for slowly decaying eigenfunction phi_40 (ablation) | Nystrom eigenfunction approximation · 1D Gaussian kernel (analytic) | eigenfunction approximation quality: n=200 -> n=1000 | Improved |
| cover 99% of variance with very wiggly kernel (sigma=0.005) (ablation) | low-rank GP approximation · 1D example | approximation error: slightly larger than 1% | Hurt |
Closest papers by meaning
- Probabilistic Shape Models of Anatomy Directly from Images (AAAI 2023 · similarity 0.86)
- Point2SSM: Learning Morphological Variations of Anatomies from Point Clouds (ICLR 2024 · similarity 0.84)
- Mixture of Probabilistic Principal Component Analyzers for Shapes from Point Sets (TPAMI 2017 · similarity 0.84)
- Procrustes Analysis with Deformations: A Closed-Form Solution by Eigenvalue Decomposition (IJCV 2022 · similarity 0.84)
- PointGMM: A Neural GMM Network for Point Clouds (CVPR 2020 · similarity 0.83)
- GOGMA: Globally-Optimal Gaussian Mixture Alignment (CVPR 2016 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
