Research map / Point clouds / Shape and surface completion
RfD-Net: Point Scene Understanding by Semantic Instance Reconstruction
Yinyu Nie, Ji Hou, Xiaoguang Han, Matthias Niessner
Check 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 |
|---|---|---|---|
| jointly train shape prediction with 3D detection (ablation) | RfD-Net with VoteNet · ScanNet v2 | mAP@0.5: 32.63 -> 35.10 | Improved |
| jointly train shape prediction with detection (ablation) | RfD-Net · ScanNet v2 semantic instance completion | mAP (mesh IoU threshold 0.5): 13.94 -> 16.90 | Improved |
| jointly train detector and shape generator (ablation) | RfD-Net · ScanNet v2 object reconstruction, 16-d | 3D IoU: 35.75 -> 37.02 | Improved |
| jointly train shape prediction with the detector (ablation) | RfD-Net with BoxNet · ScanNet v2 3D detection | mAP@0.5: 20.31 -> 22.79 | Improved |
| jointly train shape prediction with the detector (ablation) | RfD-Net with MLCVNet · ScanNet v2 3D detection | mAP@0.5: 33.40 -> 34.77 | Improved |
| jointly train shape prediction with the detector (ablation) | RfD-Net with VoteNet+DGCNN · ScanNet v2 3D detection | mAP@0.5: 29.88 -> 32.30 | Improved |
| increase input points from 20K to 80K (ablation) | RfD-Net · ScanNet v2 | 3D Detection: 33.89 -> 35.10 | Improved |
| increase input points from 20K to 80K (ablation) | RfD-Net · ScanNet v2 | Instance Completion: 14.39 -> 16.90 | Improved |
| set objectness dropout to Nd=30 rather than Nd=10 (ablation) | RfD-Net · ScanNet v2 | 3D Detection: 35.10 -> 33.37 | Hurt |
| set objectness dropout to Nd=30 rather than Nd=10 (ablation) | RfD-Net · ScanNet v2 | Instance Completion: 16.90 -> 14.34 | Hurt |
| combine proposal features and point clusters through full skip propagation, including denoising (ablation) | RfD-Net · ScanNet v2 | Instance Completion: 15.42 -> 16.90 | Improved |
Later papers that compared with this method
2 later accepted papers report RfD-Net as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Point-based Instance Completion with Scene Constraints (ICLR 2025 · better on 11 of 12 shared results)
- Point Scene Understanding via Disentangled Instance Mesh Reconstruction (ECCV 2022 · better on 9 of 10 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
- Point Scene Understanding via Disentangled Instance Mesh Reconstruction (ECCV 2022 · similarity 0.92)
- Know Your Neighbors: Improving Single-View Reconstruction via Spatial Vision-Language Reasoning (CVPR 2024 · similarity 0.89)
- RFNet-4D: Joint Object Reconstruction and Flow Estimation from 4D Point Clouds (ECCV 2022 · similarity 0.89)
- Associatively Segmenting Instances and Semantics in Point Clouds (CVPR 2019 · similarity 0.89)
- VoxDet: Rethinking 3D Semantic Scene Completion as Dense Object Detection (NeurIPS 2025 · similarity 0.89)
- A Survey and Benchmark of Automatic Surface Reconstruction From Point Clouds (TPAMI 2024 · 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.
