Research map / Probabilistic and scientific modeling / Molecules and proteins
Modeling Facial Geometry Using Compositional VAEs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace the global linear model with a compositional VAE | face geometry fitting · 3D-3D correspondences | RMSE (mm): 2.795 -> 1.320 (0.2%); 1.016 -> 0.247 (10%) | Helped |
| replace the single-level VAE with a compositional VAE | face geometry fitting · 3D-3D correspondences | RMSE (mm): 1.678 -> 1.320 (0.2%) | Helped |
| use priors on all layers z1:4 rather than only z1 (ablation) | compositional VAE · 3D-3D correspondences (0.2% constrained vertices) | RMSE (mm): 1.470 -> 1.320 | Helped |
| use only the fine-detail prior z1 rather than priors on z1:4 (ablation) | compositional VAE · 3D-3D correspondences (10% constrained vertices) | RMSE (mm): 0.717 -> 0.247 | Helped |
| replace the global linear model with a compositional VAE | face geometry fitting · 2D-3D correspondences | RMSE (mm): 4.381 -> 2.586 (0.2%); 3.302 -> 2.330 (10%) | Helped |
| use priors on all layers z1:4 rather than only z1 (ablation) | compositional VAE · 2D-3D correspondences (0.2% constrained vertices) | RMSE (mm): 2.690 -> 2.586 | Helped |
| use only the fine-detail prior z1 rather than priors on z1:4 (ablation) | compositional VAE · 2D-3D correspondences (10% constrained vertices) | RMSE (mm): 2.453 -> 2.330 | Helped |
| replace the global linear model with a compositional VAE | face geometry fitting · Depth maps | RMSE (mm): 3.908 -> 3.020 (σ² = 1); 3.953 -> 3.191 (σ² = 3) | Helped |
| use priors through z3 rather than through z4 (ablation) | compositional VAE · Depth maps (σ² = 3) | RMSE (mm): 3.226 -> 3.191 | Helped |
| use priors on all layers z1:4 rather than only z1:3 (ablation) | compositional VAE · Depth maps (σ² = 1) | RMSE (mm): 3.079 -> 3.110 | Hurt |
| use a compositional latent hierarchy instead of a single low-dimensional latent vector | face geometry representation · 3D face geometry | RMSE (mm): LINEAR 2.795 -> OURS z1:4 1.320 (3D-3D, 0.2%) | Helped |
Closest papers by meaning
- Learning Formation of Physically-Based Face Attributes (CVPR 2020 · similarity 0.87)
- Mesoscopic Facial Geometry Inference Using Deep Neural Networks (CVPR 2018 · similarity 0.87)
- 3D Shape Variational Autoencoder Latent Disentanglement via Mini-Batch Feature Swapping for Bodies and Faces (CVPR 2022 · similarity 0.86)
- Explicit Disentanglement of Appearance and Perspective in Generative Models (NeurIPS 2019 · similarity 0.86)
- Self-Supervised Multi-Level Face Model Learning for Monocular Reconstruction at Over 250 Hz (CVPR 2018 · similarity 0.86)
- FACE: A Face-based Autoregressive Representation for High-Fidelity and Efficient Mesh Generation (CVPR 2026 · similarity 0.86)
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.