Research map / 3D vision / Face and shape
Dense 3D Face Correspondence
Syed Zulqarnain Gilani, Ajmal Saeed Mian, Faisal Shafait B, Ian D. Reid
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 |
|---|---|---|---|
| add organising faces into a graph (ablation) | K3DM dense correspondence · Synthetic faces (2,246 vertices) | mean localization error (mm): 2.16 -> 1.28 | Improved |
| add keypoint detection (ablation) | K3DM dense correspondence · Synthetic faces (2,246 vertices) | mean localization error (mm): 3.06 -> 1.28 | Improved |
| add feature matching (ablation) | K3DM dense correspondence · Synthetic faces (2,246 vertices) | mean localization error (mm): 3.61 -> 1.28 | Improved |
| add keypoint detection and feature matching together (ablation) | K3DM dense correspondence · Synthetic faces (2,246 vertices) | mean localization error (mm): 4.78 -> 1.28 | Improved |
| add selecting best matches in each iteration (ablation) | K3DM dense correspondence · Synthetic faces (2,246 vertices) | mean localization error (mm): 2.61 -> 1.28 | Improved |
| use NICP variant for densification instead of level-set geodesic curves | K3DM · Synthetic faces | mean localization error (mm): 1.28 -> 1.30 | No clear effect |
| use K3DM_FR instead of cross-domain K3DM_BO on FRGCv2 landmarks | K3DM · FRGCv2 landmark localization | mean landmark error (mm): 3.8 -> 2.9 | Improved |
| use K3DM_BO instead of BFM for Bosphorus landmarks | BFM · Bosphorus landmark localization (all) | mean landmark error (mm): 4.48 -> 3.70 | Improved |
| use K3DM_BU instead of BFM for cross-domain FRGCv2 recognition | BFM · FRGCv2 rank-1 (cross domain) | rank-1 accuracy (%): 76.4% -> 80.5% | Improved |
| use K3DM_BO instead of BFM for cross-domain FRGCv2 recognition | BFM · FRGCv2 rank-1 (cross domain) | rank-1 accuracy (%): 76.4% -> 77.1% | Improved |
| use K3DM_BU instead of BFM for cross-domain Bosphorus recognition | BFM · Bosphorus rank-1 (cross domain) | rank-1 accuracy (%): 82.7% -> 90.7% | Improved |
| use K3DM_FR for UND45LR rank-1 recognition vs Smeets et al. | K3DM_FR · UND45LR | rank-1 accuracy (%): 98.3% vs 95.8% | Hurt |
Closest papers by meaning
- Dense Semantic and Topological Correspondence of 3D Faces without Landmarks (ECCV 2018 · similarity 0.91)
- Towards Fine-Grained Optimal 3D Face Dense Registration: An Iterative Dividing and Diffusing Method (IJCV 2023 · similarity 0.91)
- Boosting Local Shape Matching for Dense 3D Face Correspondence (CVPR 2019 · similarity 0.91)
- A Sparse and Locally Coherent Morphable Face Model for Dense Semantic Correspondence Across Heterogeneous 3D Faces (TPAMI 2021 · similarity 0.91)
- Functional Faces: Groupwise Dense Correspondence Using Functional Maps (CVPR 2016 · similarity 0.90)
- Learning Dense Facial Correspondences in Unconstrained Images (ICCV 2017 · similarity 0.89)
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.
