Research map / 3D vision / Face and shape
HiFace: High-Fidelity 3D Face Reconstruction by Learning Static and Dynamic Details
Zenghao Chai, Tianke Zhang, Tianyu He, Xu Tan, Tadas Baltrusaitis, HsiangTao Wu, Runnan Li, Sheng Zhao, Chun Yuan, Jiang Bian
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 |
|---|---|---|---|
| train with ground-truth labels from synthetic data (coarse shape) (ablation) | HiFace coarse · REALY frontal-view | all-region error (mm): 1.558 -> 1.297 | Improved |
| train with ground-truth labels from synthetic data (detail) (ablation) | HiFace detail · REALY frontal-view | all-region error (mm): 1.600 -> 1.275 | Improved |
| add static detail factor to the detail model (ablation) | HiFace detail · REALY frontal-view | all-region error (mm): 1.295 -> 1.275 | Improved |
| add dynamic detail factor to the detail model (ablation) | HiFace detail · REALY frontal-view | all-region error (mm): 1.292 -> 1.275 | Improved |
| add dynamic factor to the detail model (side view) (ablation) | HiFace detail · REALY side-view | all-region error (mm): 1.318 -> 1.308 | Improved |
| add detail reconstruction on top of coarse shape | HiFace · REALY frontal-view | all-region error (mm): 1.297 -> 1.275 | Improved |
| image-level-supervised detail modeling (DECA, EMOCA) added to coarse shape | DECA/EMOCA · REALY frontal-view | all-region error (mm): 2.010 -> 2.210 (DECA); 2.103 -> 2.388 (EMOCA) | Hurt |
| directly synthesize dynamic detail with U-Net instead of interpolating compressed/stretched maps (ablation) | SD-DeTail · qualitative ablation (SD-1) | visual quality | Hurt |
| replace PCA displacement bases with U-Net decoders for static and dynamic detail (ablation) | SD-DeTail · qualitative ablation (SD-2, SD-3) | visual quality | Hurt |
| add real-world data to synthetic training data (ablation) | HiFace · qualitative ablation | generalization to real images | Improved |
| add KL coefficient loss L_kl (ablation) | HiFace coarse shape · qualitative ablation | generalization to real images | Improved |
| add detail loss and age distillation loss L_kd (ablation) | HiFace detail · qualitative ablation | detail quality | Improved |
Later papers that compared with this method
2 later accepted papers report HiFace as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Monocular Avatar Reconstruction via Cascaded Diffusion Priors and UV-Space Differentiable Shading (ECCV 2026 · better on 0 of 1 shared result)
- MoSAR: Monocular Semi-Supervised Model for Avatar Reconstruction using Differentiable Shading (CVPR 2024 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Deep3D — Deep3D: Fully Automatic 2D-to-3D Video Conversion with Deep Convolutional Neural Networks (ECCV 2016) · better on 2 of 2 shared results
- EMOCA — EMOCA: Emotion Driven Monocular Face Capture and Animation (CVPR 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Hierarchical Representation Network for Accurate and Detailed Face Reconstruction From In-the-Wild Images (CVPR 2023 · similarity 0.92)
- 3D Reconstruction of “In-the-Wild” Faces in Images and Videos (TPAMI 2018 · similarity 0.91)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.91)
- On Learning 3D Face Morphable Model from In-the-Wild Images (TPAMI 2019 · similarity 0.91)
- Self-Supervised Multi-Level Face Model Learning for Monocular Reconstruction at Over 250 Hz (CVPR 2018 · similarity 0.90)
- Nonlinear 3D Face Morphable Model (CVPR 2018 · similarity 0.90)
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.
