Research map / 3D vision / Face and shape
A Sparse and Locally Coherent Morphable Face Model for Dense Semantic Correspondence Across Heterogeneous 3D Faces
Claudio Ferrari, Stefano Berretti, Pietro Pala, Alberto Del Bimbo
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 |
|---|---|---|---|
| replace nearest-neighbor correspondence with mean-point association and outlier rejection (ablation) | NRF with SLC · FRGCv2.0, Bosphorus, FaceWarehouse | landmark error | Improved |
| use SLC components instead of PCA in NRF (ablation) | NRF · FaceWarehouse landmark localization | avg landmark error (mm): 4.6 -> 4.1 | Improved |
| use SLC components instead of DL in NRF (ablation) | NRF · FaceWarehouse landmark localization | avg landmark error (mm): 4.3 -> 4.1 | Improved |
| use SLC components instead of PCA in NRF (ablation) | NRF · FRGCv2.0 landmark localization | avg landmark error (mm): 3.0 -> 2.8 | Improved |
| use SLC components instead of DL in NRF (ablation) | NRF · Bosphorus landmark localization (Expression subset) | avg landmark error (mm): 3.0 -> 3.1 | No clear effect |
| register with NRF+SLC instead of NICP before building PCA model (ablation) | HLS-3DMM · HLS-3DMM compactness (9,927 scans) | components for 90% variance: 114 -> 43 | Improved |
| use SLC components instead of PCA in registration for compactness (ablation) | HLS-3DMM · HLS-3DMM compactness (9,927 scans) | components for 90% variance: 47 -> 43 | Improved |
| reduce CoMA training set to 360 scans (2%) (ablation) | SLC · CoMA (identity split, k=512) | vertices with error below 1mm (%): 92% -> 79% | Hurt |
| reduce CoMA training set to 360 scans (2%) (ablation) | Neural3DMM · CoMA (identity split, k=512) | vertices with error below 1mm (%): drops down to 0% | Hurt |
| increase number of components k from 64 to 512 (ablation) | SLC · SLC component sparsity | zero entries in C (%): 84.4% -> 96.3% | Mixed |
| remove l2 regularization from SLC learning (l1 only) (ablation) | SLC components · component visualization | deformation smoothness | Hurt |
| use SLC instead of PCA with small k (up to 64 components) (ablation) | CoMA identity split · CoMA identity split | reconstruction error | Hurt |
Closest papers by meaning
- Large Scale 3D Morphable Models (IJCV 2017 · similarity 0.92)
- i3DMM: Deep Implicit 3D Morphable Model of Human Heads (CVPR 2021 · similarity 0.91)
- A 3D Morphable Model Learnt From 10,000 Faces (CVPR 2016 · similarity 0.91)
- Combining 3D Morphable Models: A Large Scale Face-And-Head Model (CVPR 2019 · similarity 0.91)
- ImFace++: A Sophisticated Nonlinear 3D Morphable Face Model With Implicit Neural Representations (TPAMI 2024 · similarity 0.91)
- Dense 3D Face Correspondence (TPAMI 2017 · similarity 0.91)
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.
