Research map / 3D vision / NeRF and radiance fields
SceneRF: Self-Supervised Monocular 3D Scene Reconstruction with Radiance Fields
Anh-Quan Cao, Raoul de Charette
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 reprojection loss for explicit depth optimization (ablation) | SceneRF · SemanticKITTI novel depth | delta1: 69.10 -> 75.07 | Improved |
| add reprojection loss for explicit depth optimization (ablation) | SceneRF · BundleFusion novel depth | delta1: 64.99 -> 72.71 | Improved |
| add reprojection loss to other NeRFs | VisionNeRF · SemanticKITTI novel depth | delta1: 69.11 -> 75.77 | Improved |
| add probabilistic ray sampling (PrSamp) (ablation) | SceneRF · SemanticKITTI novel depth | delta1: 71.85 -> 75.07 | Improved |
| add Spherical U-Net decoder (ablation) | SceneRF · BundleFusion novel depth | delta1: 59.77 -> 72.71 | Improved |
| add RGB loss alongside reprojection loss (ablation) | SceneRF · SemanticKITTI novel depth | delta1: 72.15 -> 75.07 | Improved |
| add depth components (reproj, PrSamp, SU-Net) on novel view synthesis (ablation) | SceneRF · SemanticKITTI novel view | LPIPS: 0.464 -> 0.476 (w/o SU-Net vs full) | Mixed |
| increase number of Gaussians or points per Gaussian beyond k=4, m=8 (ablation) | PrSamp · SemanticKITTI novel depth | delta1: 75.01 -> 74.59 (k=8, m=8) | No clear effect |
| replace PrSamp with MipNeRF360 sampling (ablation) | SceneRF · SemanticKITTI and BundleFusion novel depth | delta1/delta2/delta3: -6/-3/-2 on SemanticKITTI | Hurt |
| add fused novel depth views reconstruction scheme (ablation) | SceneRF · BundleFusion scene reconstruction | IoU: 17.33 -> 20.16 | Improved |
| use self-supervised image supervision instead of depth supervision for scene reconstruction | SceneRF vs MonoScene (depth) · SemanticKITTI scene reconstruction | IoU: 13.53 -> 13.84 | Improved |
| use spherical 120-degree decoder instead of planar decoders (ablation) | SU-Net · SemanticKITTI novel depth | delta1: 73.78 -> 75.01 | Improved |
Later papers that compared with this method
3 later accepted papers report SceneRF as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Hi-Gaussian: Hierarchical Gaussians under Normalized Spherical Projection for Single-View 3D Reconstruction (ICCV 2025 · better on 6 of 6 shared results)
- SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction (CVPR 2024 · better on 3 of 3 shared results)
- Novel View Synthesis with View-Dependent Effects from a Single Image (CVPR 2024 · 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
- S$^3$-NeRF: Neural Reflectance Field from Shading and Shadow under a Single Viewpoint (NeurIPS 2022 · similarity 0.90)
- AltNeRF: Learning Robust Neural Radiance Field via Alternating Depth-Pose Optimization (AAAI 2024 · similarity 0.89)
- DistillNeRF: Perceiving 3D Scenes from Single-Glance Images by Distilling Neural Fields and Foundation Model Features (NeurIPS 2024 · similarity 0.88)
- NeRFPrior: Learning Neural Radiance Field as a Prior for Indoor Scene Reconstruction (CVPR 2025 · similarity 0.88)
- 3D Reconstruction with Generalizable Neural Fields using Scene Priors (ICLR 2024 · similarity 0.87)
- BLiRF: Bandlimited Radiance Fields for Dynamic Scene Modeling (AAAI 2024 · 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.
