Research map / 3D vision / Face and shape
Towards Metrical Reconstruction of Human Faces
Wojciech Zielonka, Timo Bolkart, Justus Thies
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 ArcFace instead of FaceNet as identity encoder (ablation) | MICA · Stirling (metrical, HQ) | mean error (mm): 1.65 -> 1.52 | Improved |
| fine-tune last ResNet layer (L4) of ArcFace (ablation) | MICA frozen ArcFace · Stirling (metrical, HQ) | mean error (mm): 1.52 -> 1.35 | Improved |
| fine-tune all ArcFace layers instead of only L4 (ablation) | MICA · Stirling (metrical, HQ) | mean error (mm): 1.35 -> 1.42 | Hurt |
| fine-tune DECA encoder fully on unified dataset (ablation) | DECA · Stirling (metrical, HQ) | mean error (mm): 1.58 -> 1.82 | Hurt |
| use ArcFace features instead of fine-tuned DECA encoder (ablation) | DECA fine-tuned · Stirling (metrical, HQ) | mean error (mm): 1.82 -> 1.35 | Improved |
| increase FLAME principal components from 50 to 300 (ablation) | MICA decoder · Stirling (metrical, HQ) | mean error (mm): 1.41 -> 1.35 | Improved |
| add LYHM dataset to training data (ablation) | MICA · Stirling (metrical, HQ) | mean error (mm): 1.43 -> 1.35 | Improved |
| add LYHM dataset to training data (ablation) | MICA · Stirling (metrical, LQ) | mean error (mm): 1.51 -> 1.46 | Improved |
| add FaceWarehouse dataset to training data (ablation) | MICA · Stirling (metrical, HQ) | mean error (mm): 1.36 -> 1.35 | No clear effect |
| use FLAME decoder instead of model-free SIREN decoder (ablation) | MICA · Stirling (metrical, HQ) | mean error (mm): 1.39 -> 1.35 | No clear effect |
| evaluate with scale optimization (non-metrical) instead of rigid alignment | Average FLAME face · NoW | mean error (mm): 1.92 -> 1.53 | Improved |
| replace DECA with MICA shape prediction | DECA · NoW (metrical) | mean error (mm): 1.80 -> 1.37 | Improved |
Closest papers by meaning
- Large Pose 3D Face Reconstruction From a Single Image via Direct Volumetric CNN Regression (ICCV 2017 · similarity 0.89)
- Face Anything: 4D Face Reconstruction from Any Image Sequence (ECCV 2026 · similarity 0.88)
- MMFace: A Multi-Metric Regression Network for Unconstrained Face Reconstruction (CVPR 2019 · similarity 0.88)
- A Simple, Fast and Highly-Accurate Algorithm to Recover 3D Shape from 2D Landmarks on a Single Image (TPAMI 2017 · similarity 0.87)
- Learning Detailed Face Reconstruction From a Single Image (CVPR 2017 · similarity 0.87)
- FaceScape: 3D Facial Dataset and Benchmark for Single-View 3D Face Reconstruction (TPAMI 2023 · similarity 0.87)
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.
