Research map / 3D vision / 3D generation
FaceLift: Learning Generalizable Single Image 3D Face Reconstruction from Synthetic Heads
Weijie Lyu, Yi Zhou, Ming-Hsuan Yang, Zhixin Shu
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 input view reconstruction to multi-view diffusion training (ablation) | FaceLift · Cafca | ArcFace distance: 0.2367 -> 0.1573 | Improved |
| add input view reconstruction to multi-view diffusion training (ablation) | FaceLift · Cafca | DreamSim: 0.1438 -> 0.1096 | Improved |
| add input view reconstruction to multi-view diffusion training (ablation) | FaceLift · Cafca | PSNR: 16.02 -> 16.61 | Improved |
| add multi-view attention (ablation) | FaceLift · Cafca | DreamSim: 0.1552 -> 0.1096 | Improved |
| add multi-view attention (ablation) | FaceLift · Cafca | ArcFace distance: 0.2126 -> 0.1573 | Improved |
| replace LGM input views with FaceLift multi-view diffusion outputs | LGM · Cafca | ArcFace distance: 0.2557 -> 0.1767 | Improved |
| replace LGM input views with FaceLift multi-view diffusion outputs | LGM · Cafca | PSNR: 16.52 -> 14.13 | Hurt |
| replace LGM input views with FaceLift multi-view diffusion outputs | LGM · Ava-256 | PSNR: 14.05 -> 15.24 | Improved |
| replace LGM reconstructor with FaceLift transformer Gaussian reconstructor | Our MV + LGM · Cafca | DreamSim: 0.1282 -> 0.1096 | Improved |
| fine-tune LGM on synthetic head data | LGM · Cafca | visual quality | Hurt |
| use six views instead of four views (ablation) | FaceLift reconstructor · in-the-wild images | visual quality | Improved |
| use eight views instead of six views (ablation) | FaceLift · in-the-wild images | visual quality | No clear effect |
Later papers that compared with this method
3 later accepted papers report FaceLift as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Feed-forward Gaussian Registration for Head Avatar Creation and Editing (CVPR 2026 · better on 3 of 3 shared results)
- PercHead: Perceptual Head Model for Single-Image 3D Head Reconstruction & Editing (CVPR 2026 · better on 2 of 2 shared results)
- VoluMe - Authentic 3D Video Calls from Live Gaussian Splat Prediction (ICCV 2025 · better on 4 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PanoHead — PanoHead: Geometry-Aware 3D Full-Head Synthesis in 360deg (CVPR 2023) · better on 3 of 3 shared results
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
- Large-Scale High-Quality 3D Gaussian Head Reconstruction from Multi-View Captures (ECCV 2026 · similarity 0.89)
- InverseFaceNet: Deep Monocular Inverse Face Rendering (CVPR 2018 · similarity 0.89)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.88)
- Self-Supervised Multi-Level Face Model Learning for Monocular Reconstruction at Over 250 Hz (CVPR 2018 · similarity 0.88)
- PSHuman: Photorealistic Single-image 3D Human Reconstruction using Cross-Scale Multiview Diffusion and Explicit Remeshing (CVPR 2025 · similarity 0.88)
- Synthetic Prior for Few-Shot Drivable Head Avatar Inversion (CVPR 2025 · 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.
