Research map / 3D vision / Face and shape
Learning Priors for Semantic 3D Reconstruction
Ian Cherabier, Johannes L. Schonberger, Martin R. Oswald, Marc Pollefeys, Andreas Geiger
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 unrolled variational regularization (50 iterations) to the network (ablation) | Ours-300 · ScanNet (156 test scenes, weak data cost) | overall accuracy: 97.3 -> 98.7 | Improved |
| add unrolled variational regularization (50 iterations) to the network (ablation) | Ours-300 · ScanNet (156 test scenes, weak data cost) | freespace accuracy: 97.6 -> 98.6 | Improved |
| add unrolled variational regularization (50 iterations) to the network (ablation) | Ours-300 · ScanNet (156 test scenes, weak data cost) | occupied accuracy: 92.3 -> 94.4 | Improved |
| add unrolled variational regularization (50 iterations) to the network (ablation) | Ours-300 · ScanNet (156 test scenes, weak data cost) | semantic accuracy: 90.2 -> 91.5 | Improved |
| train on 312 scenes instead of only 5 scenes (ablation) | Ours · ScanNet (156 test scenes, weak data cost) | overall accuracy: 96.7 -> 98.7 | Improved |
| use learned multi-scale network instead of TV-L1 at 50 iterations | Ours · ScanNet (156 test scenes, weak data cost) | overall accuracy: 92.8 -> 98.7 | Improved |
| use learned network (50 it.) instead of TV-L1 run to convergence (500 it.) | Ours · ScanNet (156 test scenes, weak data cost) | overall accuracy: 95.8 -> 98.7 | Improved |
| use multi-scale architecture instead of naive coarse-to-fine network (ablation) | Ours · ScanNet (156 test scenes, weak data cost) | overall accuracy: 21.0 -> 98.7 | Improved |
| add multi-scale (3 levels) to the unrolled network (ablation) | Ours-300 · ScanNet (156 test scenes, weak data cost) | overall accuracy: 98.7 -> 98.7 | No clear effect |
| increase TV-L1 iterations from 50 to 500 | TV-L1 · ScanNet (156 test scenes, weak data cost) | overall accuracy: 92.8 -> 95.8 | Improved |
| replace Wulffshape hand-crafted regularizer with learned priors | Haene et al. [12] · Haene et al. outdoor scenes (Castle, South Building) | iterations: 2750 -> 50 | Improved |
Closest papers by meaning
- Large-Scale Semantic 3D Reconstruction: An Adaptive Multi-Resolution Model for Multi-Class Volumetric Labeling (CVPR 2016 · similarity 0.89)
- Semantically Informed Multiview Surface Refinement (ICCV 2017 · similarity 0.88)
- SceneCode: Monocular Dense Semantic Reconstruction Using Learned Encoded Scene Representations (CVPR 2019 · similarity 0.88)
- Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation (NeurIPS 2026 · similarity 0.87)
- MESC-3D:Mining Effective Semantic Cues for 3D Reconstruction from a Single Image (CVPR 2025 · similarity 0.86)
- WorDepth: Variational Language Prior for Monocular Depth Estimation (CVPR 2024 · 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.
