Research map / 3D vision / Face and shape
H3D-Net: Few-Shot High-Fidelity 3D Head Reconstruction
Eduard Ramon, Gil Triginer, Janna Escur, Albert Pumarola, Jaime Garcia, Xavier Giró-i-Nieto, Francesc Moreno-Noguer
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 a small learned head prior (500 subjects) as initialization (ablation) | H3D-Net · H3DS (3 views, face) | mean distance (mm): 4.04 -> 2.68 | Improved |
| add a small learned head prior (500 subjects) as initialization (ablation) | H3D-Net · H3DS (3 views, head) | mean distance (mm): 16.68 -> 17.08 | Hurt |
| scale prior training set from 500 to 10,000 subjects (ablation) | H3D-Net · H3DS (3 views, face) | mean distance (mm): 2.68 -> 1.90 | Improved |
| scale prior training set from 500 to 10,000 subjects (ablation) | H3D-Net · H3DS (3 views, head) | mean distance (mm): 17.08 -> 14.59 | Improved |
| add the two-step optimization schedule (freeze decoder, then unfreeze) (ablation) | H3D-Net · H3DS (3 views, face) | mean distance (mm): 1.90 -> 1.49 | Improved |
| add the two-step optimization schedule (freeze decoder, then unfreeze) (ablation) | H3D-Net · H3DS (3 views, head) | mean distance (mm): 14.59 -> 12.76 | Improved |
| replace IDR with H3D-Net (learned prior) | H3D-Net · 3DFAW (3 views, face) | surface error (mm): 3.92 -> 1.37 | Improved |
| replace IDR with H3D-Net (learned prior) | H3D-Net · H3DS (3 views, head) | surface error (mm): 17.04 -> 12.76 | Improved |
| replace IDR with H3D-Net (learned prior) | H3D-Net · H3DS (32 views, face) | surface error (mm): 1.39 -> 1.21 | Improved |
| replace IDR with H3D-Net (learned prior) | H3D-Net · H3DS (32 views, head) | surface error (mm): 5.86 -> 4.90 | Improved |
| replace IDR with H3D-Net (learned prior) | H3D-Net · H3DS (32 views) | convergence time: 20x faster for the facial region, 4x for the full head | Improved |
| replace IDR with H3D-Net (learned prior) | H3D-Net · H3DS (4 views, head) | surface error (mm): 8.04 -> 7.95 | Improved |
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.
- MVFNet — MVF-Net: Multi-View 3D Face Morphable Model Regression (CVPR 2019) · better on 2 of 2 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
- Implicit Shape and Appearance Priors for Few-Shot Full Head Reconstruction (TPAMI 2025 · similarity 0.96)
- Skullptor: High Fidelity 3D Head Reconstruction in Seconds with Multi-View Normal Prediction (CVPR 2026 · similarity 0.90)
- Deformable Model-Driven Neural Rendering for High-Fidelity 3D Reconstruction of Human Heads Under Low-View Settings (ICCV 2023 · similarity 0.90)
- Deep 3D Portrait From a Single Image (CVPR 2020 · similarity 0.89)
- Coordinates Are NOT Lonely - Codebook Prior Helps Implicit Neural 3D representations (NeurIPS 2022 · similarity 0.88)
- One-2-3-45: Any Single Image to 3D Mesh in 45 Seconds without Per-Shape Optimization (NeurIPS 2023 · similarity 0.88)
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.
