Research map / 3D vision / Depth and stereo
Learning Visibility Field for Detailed 3D Human Reconstruction and Relighting
Ruichen Zheng, Peng Li, Haoqian Wang, Tao Yu
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 visibility-guided feature aggregation instead of averaging (ablation) | Sparse-view human reconstruction · Feature aggregation ablation | F-score (0.5%): 0.696 -> 0.736 | Improved |
| Use visibility-guided feature aggregation instead of self-attention (ablation) | Sparse-view human reconstruction · Feature aggregation ablation | F-score (0.5%): 0.710 -> 0.736 | Improved |
| Use visibility-guided feature aggregation instead of averaging (ablation) | Sparse-view human reconstruction · Feature aggregation ablation | PSNR: 17.933 -> 23.436 | Improved |
| Use visibility-guided feature aggregation instead of self-attention (ablation) | Sparse-view human reconstruction · Feature aggregation ablation | PSNR: 19.148 -> 23.436 | Improved |
| Use the 3D geometric feature alongside 2D image features for visibility prediction (ablation) | Visibility field · Visibility prediction | visibility accuracy | Improved |
| Add TransferLoss to emphasize near-surface, normal-facing visibility alignment (ablation) | Visibility field · Synthesized testset | near-surface visibility prediction accuracy: 27.5 % -> 88.4 % | Improved |
| Add TransferLoss to emphasize near-surface, normal-facing visibility alignment (ablation) | Visibility field · Real-captured data | near-surface visibility prediction accuracy: 20.8 % -> 84.7 % | Improved |
| Increase the number of views while using visibility-guided aggregation (ablation) | Sparse-view human reconstruction · Different view numbers | geometry quality | 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.
- Function4D — Function4D: Real-Time Human Volumetric Capture From Very Sparse Consumer RGBD Sensors (CVPR 2021) · 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
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.89)
- ReCap: Better Gaussian Relighting with Cross-Environment Captures (CVPR 2025 · similarity 0.87)
- Relightable and Animatable Neural Avatar from Sparse-View Video (CVPR 2024 · similarity 0.87)
- MetaGS: A Meta-Learned Gaussian-Phong Model for Out-of-Distribution 3D Scene Relighting (NeurIPS 2025 · similarity 0.87)
- Sparse Views Near Light: A Practical Paradigm for Uncalibrated Point-light Photometric Stereo (CVPR 2024 · similarity 0.86)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.86)
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.
