Research map / 3D vision / Face and shape
Surface-Centric Modeling for High-Fidelity Generalizable Neural Surface Reconstruction
Rui Peng, Shihe Shen, Kaiqiang Xiong, Huachen Gao, Jianbo Jiao, Xiaodong Gu, Ronggang Wang
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 end-to-end region sparsification (ablation) | SuRF baseline · DTU ablation | mean chamfer distance: 1.70 -> 1.02 | Improved |
| use multi-stage training instead of end-to-end sparsification (ablation) | SuRF baseline · DTU ablation | mean chamfer distance: 1.70 -> 1.52 | Improved |
| increase volume and image resolution (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.12 -> 1.02 | Improved |
| increase number of scales from 1 to 4 (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.38 -> 1.11 | Improved |
| increase number of scales from 4 to 5 (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.11 -> 1.13 | Hurt |
| add cross-scale fusion (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.15 -> 1.11 | Improved |
| add surface sampling (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.13 -> 1.11 | Improved |
| add pseudo point cloud loss (beta 1.0) (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.17 -> 1.11 | Improved |
| weight matching field loss increasing toward fine scales (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.25 -> 1.11 | Improved |
| change surface region range ratios (ablation) | SuRF · DTU ablation | mean chamfer distance: 1.10 / 1.11 / 1.12 | No clear effect |
| learn surface region jointly with matching field instead of using RC-MVSNet region (ablation) | SuRF · DTU | mean chamfer distance: 1.13 -> 1.05 | Improved |
| use SuRF instead of C2F2NeuS in memory with more views | C2F2NeuS · 9 input views | memory: 28G -> 3.1G | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- VolRecon — VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction (CVPR 2023) · better on 1 of 1 shared result
- ReTR — ReTR: Modeling Rendering Via Transformer for Generalizable Neural Surface Reconstruction (NeurIPS 2023) · better on 1 of 1 shared result
- GenS — GenS: Generalizable Neural Surface Reconstruction from Multi-View Images (NeurIPS 2023) · better on 1 of 1 shared result
- RC-MVSNet — RC-MVSNet: Unsupervised Multi-View Stereo with Neural Rendering (ECCV 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
- NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views (AAAI 2024 · similarity 0.89)
- GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction (NeurIPS 2025 · similarity 0.89)
- SC-NeuS: Consistent Neural Surface Reconstruction from Sparse and Noisy Views (AAAI 2024 · similarity 0.88)
- SurfelSplat: Learning Efficient and Generalizable Gaussian Surfel Representations for Sparse-View Surface Reconstruction (NeurIPS 2025 · similarity 0.88)
- Flatten Anything: Unsupervised Neural Surface Parameterization (NeurIPS 2024 · similarity 0.87)
- Sparis: Neural Implicit Surface Reconstruction of Indoor Scenes from Sparse Views (AAAI 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.
