Research map / 3D vision / Depth and stereo
From None to All: Self-Supervised 3D Reconstruction via Novel View Synthesis
Ranran Huang, Weixun Luo, Ye Mao, Krystian Mikolajczyk
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 NAS3R instead of SelfSplat or SPFSplat* for self-supervised 2-view NVS | NAS3R · RE10K 2-view NVS | PSNR: 21.306 -> 23.130 vs SPFSplat* | Improved |
| use NAS3R instead of SPFSplat* for pose estimation | NAS3R · RE10K 2-view pose | overall AUC@10: 23.9 -> 51.0 | Improved |
| use depth-based local-to-global Gaussian formulation without pretrained priors | NAS3R · BlendedMVS depth | rel: 0.255 -> 0.206 vs SPFSplat* | Improved |
| train on RE10K+DL3DV instead of RE10K only (ablation) | NAS3R · ScanNet++ pose AUC@10 | AUC@10: 19.2 -> 24.5 | Improved |
| train on DL3DV instead of RE10K only (ablation) | NAS3R · ScanNet++ pose AUC@10 | AUC@10: 19.2 -> 19.1 | No clear effect |
| increase number of context views from 2 to 10 (ablation) | NAS3R · RE10K NVS | PSNR: 23.130 -> 27.093 | Improved |
| initialize with MASt3R or VGGT pretrained priors (ablation) | NAS3R · RE10K 2-view NVS | PSNR: 23.130 -> 25.888 (VGGT) | Improved |
| provide GT intrinsics to the backbone (NAS3R-I) (ablation) | NAS3R (VGGT) · RE10K 2-view pose | AUC@5: 62.4 -> 67.7 | Improved |
| initialize supervised fine-tuning from NAS3R self-supervised weights (ablation) | BlendedMVS supervised training · BlendedMVS depth | rel: 0.177 -> 0.119 | Improved |
| self-supervised NAS3R versus supervised-from-scratch training on BlendedMVS pose (ablation) | NAS3R · BlendedMVS pose | AUC@5: 10.7 -> 29.8 | Improved |
| NAS3R(VGGT) versus NAS3R(MASt3R) with priors (ablation) | NAS3R · RE10K 2-view pose | AUC@5: 63.8 (MASt3R) vs 62.4 (VGGT) | No clear effect |
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.
- SelfSplat — SelfSplat: Pose-Free and 3D Prior-Free Generalizable 3D Gaussian Splatting (CVPR 2025) · better on 7 of 7 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
- RayZer: A Self-supervised Large View Synthesis Model (ICCV 2025 · similarity 0.88)
- Dens3R: A Foundation Model for 3D Geometry Prediction (ICLR 2026 · similarity 0.88)
- Selfi: Self-improving Reconstruction Engine via 3D Geometric Feature Alignment (CVPR 2026 · similarity 0.88)
- Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priors (CVPR 2025 · similarity 0.87)
- Fin3R: Fine-tuning Feed-forward 3D Reconstruction Models via Monocular Knowledge Distillation (NeurIPS 2025 · similarity 0.86)
- SelfSplat: Pose-Free and 3D Prior-Free Generalizable 3D Gaussian Splatting (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.
