Research map / 3D vision / 3D generation
Robust Shape Fitting for 3D Scene Abstraction
Florian Kluger, Eric Brachmann, Michael Ying Yang, Bodo Rosenhahn
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 occlusion-aware inlier counting at inference (ablation) | cuboid fitting · NYU validation, depth input | OA-L2 AUC@20cm (%): 6.2 -> 56.3 | Improved |
| add occlusion-aware inlier counting at inference (ablation) | cuboid fitting · NYU validation | mean OA distance covered points (cm): 135.1 -> 6.4 (roughly 1.3 m -> 6.4 cm) | Improved |
| add occlusion-aware inlier counting during training (sampling and loss) (ablation) | cuboid fitting · NYU validation | AUC: 0.2 to 1.0 percentage points | Improved |
| add occlusion-aware inlier counting to Sequential RANSAC | Sequential RANSAC · NYU depth input | mean OA distance covered points (cm): 139.3 -> 16.5 | Improved |
| use minimal set size C=6 instead of C=9 (ablation) | cuboid fitting · NYU validation | coverage and AUC at fixed hypotheses | Improved |
| use minimal set size C=6 instead of C=9 (ablation) | cuboid fitting · NYU validation | mean OA distance covered points | No clear effect |
| use neural solver instead of numerical solver | cuboid fitting · NYU depth input | computation time (s): 7.63 -> 0.87 | Improved |
| use neural solver instead of numerical solver | cuboid fitting · NYU depth input | AUC@20cm (%): 60.1 -> 49.4 | Hurt |
| use neural solver instead of numerical solver | cuboid fitting · NYU depth input | number of primitives: 6.8 -> 3.9 | Improved |
| add EM refinement (ablation) | numerical cuboid solver · NYU depth input | AUC@5cm (%): 29.9 -> 35.1 | Improved |
| add EM refinement (ablation) | numerical cuboid solver · NYU depth input | computation time (s): 7.63 -> 9.02 | Hurt |
| add regularisation losses during end-to-end fine-tuning (ablation) | full pipeline · NYU | performance | Hurt |
Closest papers by meaning
- Cuboids Revisited: Learning Robust 3D Shape Fitting to Single RGB Images (CVPR 2021 · similarity 0.98)
- Structured Prediction of Unobserved Voxels From a Single Depth Image (CVPR 2016 · similarity 0.88)
- Im2Struct: Recovering 3D Shape Structure From a Single RGB Image (CVPR 2018 · similarity 0.88)
- Holistic 3D Scene Understanding From a Single Image With Implicit Representation (CVPR 2021 · similarity 0.88)
- Neural Parts: Learning Expressive 3D Shape Abstractions With Invertible Neural Networks (CVPR 2021 · similarity 0.88)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 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.
