Research map / 3D vision / Face and shape
S-VolSDF: Sparse Multi-View Stereo Regularization of Neural Implicit Surfaces
Haoyu Wu, Alexandros Graikos, Dimitris Samaras
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 |
|---|---|---|---|
| use MVS probability volumes with soft consistency only, without extra optimization (ablation) | S-VolSDF on CasMVSNet · DTU 3-view | Chamfer distance (mm): 1.920 -> 1.711 | Improved |
| add generalized cross entropy (noise-tolerant) weight loss instead of cross entropy (ablation) | S-VolSDF · DTU 3-view | Chamfer distance (mm): 1.534 -> 1.320 | Improved |
| use weight loss instead of MSE to filtered MVS depth (ablation) | S-VolSDF · DTU 3-view | Chamfer distance (mm): 1.792 -> 1.320 | Improved |
| use probability volumes instead of argmax depth maps as MVS input (ablation) | S-VolSDF · DTU 3-view | Chamfer distance (mm): 1.543 -> 1.320 | Improved |
| plug S-VolSDF into different MVS models (ablation) | TransMVSNet, UCSNet, CasMVSNet · DTU 3-view | Chamfer distance (mm): 2.92 -> 1.80; 2.20 -> 1.52; 1.92 -> 1.32 | Improved |
| add image interpolation to the rendering process (Ours_IR) (ablation) | S-VolSDF · DTU novel view synthesis | LPIPS: 0.321 -> 0.157 | Improved |
| apply the method to MVS stages 1, 1+2, 1+2+3 instead of stage 1 only (ablation) | S-VolSDF CasMVSNet · DTU 3-view | Chamfer distance (mm): 1.320 -> 1.312 -> 1.309 | No clear effect |
| vary loss parameter q across 0.2 to 0.8 (ablation) | weight loss · DTU 3-view | Chamfer distance (mm): overall error 1.32-1.44 | No clear effect |
| increase input views from 3 to 9 (ablation) | S-VolSDF versus VolSDF · DTU novel view synthesis PSNR | PSNR: 3 views 20.21 vs 16.99; 9 views 22.98 vs 23.04 | Mixed |
| optimize with S-VolSDF versus fully trained VolSDF | S-VolSDF · DTU training time | optimization time: 10-15 minutes versus 4-10 hours | Improved |
| lower the image resolution for three given views (ablation) | S-VolSDF versus CasMVSNet and VolSDF · DTU | Chamfer distance (mm): 1.32, 1.33, 1.33 across low, mid, high | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- GeoNeRF — GeoNeRF: Generalizing NeRF With Geometry Priors (CVPR 2022) · better on 5 of 5 shared results
- TransMVSNet — TransMVSNet: Global Context-Aware Multi-View Stereo Network With Transformers (CVPR 2022) · better on 1 of 1 shared result
- NeuS — NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction (NeurIPS 2021) · 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
- SurfaceNet+: An End-to-end 3D Neural Network for Very Sparse Multi-View Stereopsis (TPAMI 2020 · similarity 0.90)
- MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstruction (NeurIPS 2022 · similarity 0.89)
- Just a Few Points Are All You Need for Multi-View Stereo: A Novel Semi-Supervised Learning Method for Multi-View Stereo (ICCV 2021 · similarity 0.89)
- RC-MVSNet: Unsupervised Multi-View Stereo with Neural Rendering (ECCV 2022 · similarity 0.89)
- Stereo Radiance Fields (SRF): Learning View Synthesis for Sparse Views of Novel Scenes (CVPR 2021 · similarity 0.89)
- Geo-Neus: Geometry-Consistent Neural Implicit Surfaces Learning for Multi-view Reconstruction (NeurIPS 2022 · similarity 0.89)
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.
