Research map / 3D vision / 3D generation
Convex Decomposition of Indoor Scenes
Vaibhav Vavilala, David Forsyth
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 |
|---|---|---|---|
| initialize refinement with network prediction instead of random start (ablation) | convex decomposition refinement · NYUv2 test | AbsRel: 0.447 -> 0.163 | Improved |
| add refinement (polishing) after the network prediction (ablation) | network-only decomposition · NYUv2 test | AbsRel: 0.179 -> 0.163 | Improved |
| add backward-selection pruning during refinement (ablation) | refinement · NYUv2 test, non-oracle | AbsRel: 0.166 -> 0.144 (n_used 24.0 -> 14.4) | Improved |
| use GT depth in refinement instead of inferred depth (ablation) | refinement · NYUv2 test | AbsRel: 0.144 -> 0.098 | Improved |
| use GT segmentation instead of inferred segmentation in refinement (ablation) | refinement · NYUv2 test | AbsRel: 0.098 -> 0.098 | No clear effect |
| add entropy segmentation loss (ablation) | convex decomposition · NYUv2 test | Seg accuracy: 0.629 -> 0.615 | No clear effect |
| add Manhattan world losses (ablation) | convex decomposition · NYUv2 test, non-oracle | AbsRel: 0.189 -> 0.144 | Improved |
| split each convex into 8 pieces before refinement (ablation) | refinement · NYUv2 test, non-oracle | AbsRel / Seg accuracy: AbsRel 0.144 -> 0.183; Seg_Acc 0.615 -> 0.630 | Mixed |
| increase initial convex count beyond 24 (36, 48, 60) (ablation) | convex decomposition · NYUv2 test, non-oracle | AbsRel: 0.144 -> 0.198 (36) | Hurt |
| reduce initial convex count to 12 (ablation) | convex decomposition · NYUv2 test, non-oracle | AbsRel: 0.144 -> 0.185 | Hurt |
| add volume ordering loss (ablation) | convex decomposition · NYUv2 test, non-oracle | AbsRel: 0.144 -> 0.144 | No clear effect |
| keep localization loss (ablation) | convex decomposition · NYUv2 test, non-oracle | AbsRel: 0.499 -> 0.144 | Improved |
Closest papers by meaning
- Toward Practical Monocular Indoor Depth Estimation (CVPR 2022 · similarity 0.87)
- Complete 3D Scene Parsing from an RGBD Image (IJCV 2018 · similarity 0.86)
- Automatic 3D Indoor Scene Modeling From Single Panorama (CVPR 2018 · similarity 0.86)
- Locality-Sensitive Deconvolution Networks With Gated Fusion for RGB-D Indoor Semantic Segmentation (CVPR 2017 · similarity 0.86)
- What Synthesis Is Missing: Depth Adaptation Integrated With Weak Supervision for Indoor Scene Parsing (ICCV 2019 · similarity 0.86)
- SparseGNV: Generating Novel Views of Indoor Scenes with Sparse RGB-D Images (AAAI 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.
