Research map / 3D vision / 3D generation
Complete 3D Scene Parsing from an RGBD Image
Chuhang Zou, Ruiqi Guo, Zhizhong Li, Derek Hoiem
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 support height feature to region classification (ablation) | Region classification CNN · 81-class region classification (GT regions) | avg per class accuracy (%): 43.7 -> 44.7 | Improved |
| Add support height feature, regions with occlusion ratio above 0.5 (ablation) | Region classification CNN · Region classification, occlusion > 0.5 | accuracy (%): 35.7 -> 38.5 | Improved |
| Add support height feature, regions with occlusion ratio below 0.5 (ablation) | Region classification CNN · Region classification, occlusion < 0.5 | accuracy (%): 44.2 -> 45.8 | Improved |
| Add support height to the scene parsing pipeline (ablation) | Full pipeline · NYUd v2 84-class semantic segmentation | avg class accuracy: 10.51 -> 10.77 | Improved |
| Add support height to the scene parsing pipeline (ablation) | Full pipeline · NYUd v2 84-class instance segmentation | mean coverage unweighted: 34.34 -> 33.92 | Hurt |
| Use ground truth support height instead of predicted (ablation) | Full pipeline · NYUd v2 84-class semantic segmentation | avg class accuracy: 10.77 -> 11.28 | Improved |
| Use ground truth regions instead of automatic region proposals (ablation) | Full pipeline · NYUd v2 84-class semantic segmentation | avg class accuracy: 10.77 -> 16.63 | Improved |
| Use learned retrieval of similar shapes instead of Guo et al. | Shape retrieval · NYUd v2 shape retrieval | avg class accuracy top 1 (%): 11.97 -> 41.56 | Improved |
| Model layout in 3D with full pipeline instead of Guo et al. | Layout estimation · NYUd v2 layout depth | overall depth error: 0.166 -> 0.150 | Improved |
| Compare against Deep Sliding Shapes on 3D amodal detection | Full pipeline · NYUd v2 3D amodal detection | chair mAP (%): 61.1 -> 3.1 | Hurt |
| Compare against SSCNet on scene completion | Full pipeline · NYUd v2 scene completion | IoU: 55.1 -> 49.4 | Hurt |
Closest papers by meaning
- Holistic 3D Scene Parsing and Reconstruction from a Single RGB Image (ECCV 2018 · similarity 0.89)
- Learning 3D Semantic Scene Graphs From 3D Indoor Reconstructions (CVPR 2020 · similarity 0.89)
- ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes (CVPR 2017 · similarity 0.89)
- Factoring Shape, Pose, and Layout From the 2D Image of a 3D Scene (CVPR 2018 · similarity 0.89)
- 3D-to-2D Distillation for Indoor Scene Parsing (CVPR 2021 · similarity 0.89)
- Total3DUnderstanding: Joint Layout, Object Pose and Mesh Reconstruction for Indoor Scenes From a Single Image (CVPR 2020 · similarity 0.89)
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.
