Research map / 3D vision / Face and shape
Deformable Model-Driven Neural Rendering for High-Fidelity 3D Reconstruction of Human Heads Under Low-View Settings
Baixin Xu, Jiarui Zhang, Kwan-Yee Lin, Chen Qian, Ying He
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 displacement field in Stage 2 (ablation) | Ours · PR-Senior (10 views) | Chamfer distance (1e-4): 1.163 -> 0.792 | Improved |
| add displacement field in Stage 2 (ablation) | Ours · PR-Young (10 views) | Chamfer distance (1e-4): 1.842 -> 0.723 | Improved |
| add TV regularization on displacement (ablation) | Ours · Model 619 qualitative (10 views) | visual noise | Improved |
| two-stage training with pre-trained multi-person template instead of single-stage simultaneous learning (HF-NeuS style) | Ours vs HF-NeuS · PR-Senior/PR-Young, 10 views | reconstruction failures: HF-NeuS failed on 19 of 30 persons at 10 views | Improved |
| geometry decomposition with template, deformation and displacement vs NeuS | Ours · PR-Senior (10 views) | Chamfer distance (1e-4): 1.795 -> 0.785 | Improved |
| increase input views to 20 | Ours vs baselines · PR-Senior/PR-Young | Chamfer distance gap: PR-Young 20 views: 0.818 vs HF-NeuS 0.831 | No clear effect |
| proposed method vs baselines on training-view PSNR | Ours · PR-Senior (10 views) | PSNR training views: 34.81 (HF-NeuS) vs 33.78 (Ours) | Hurt |
| proposed method vs baselines on novel-view PSNR | Ours · PR-Senior (10 views) | PSNR novel views: 21.24 (HF-NeuS) -> 27.89 | Improved |
| use pre-trained template on 30 identities instead of template trained on 3 new identities (ablation) | Ours (unseen identities, 5 views) · Identities 552, 555, 598 | Chamfer distance (1e-4): 1.815 -> 1.197 (id 552) | Improved |
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.
- HF-NeuS — HF-NeuS: Improved Surface Reconstruction Using High-Frequency Details (NeurIPS 2022) · better on 6 of 7 shared results
- NeuS — NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction (NeurIPS 2021) · better on 6 of 6 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
- H3D-Net: Few-Shot High-Fidelity 3D Head Reconstruction (ICCV 2021 · similarity 0.90)
- Implicit Shape and Appearance Priors for Few-Shot Full Head Reconstruction (TPAMI 2025 · similarity 0.89)
- Implicit Neural Representations With Structured Latent Codes for Human Body Modeling (TPAMI 2023 · similarity 0.88)
- Implicit Neural Head Synthesis via Controllable Local Deformation Fields (CVPR 2023 · similarity 0.88)
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.87)
- Real-time Free-view Human Rendering from Sparse-view RGB Videos using Double Unprojected Textures (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.
