Research map / Point clouds / Segmentation and detection
Learning Inner-Group Relations on Point Clouds
Haoxi Ran, Wei Zhuo, Jun Liu, Li Lu
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 cross-channel attention in place of vanilla attention (ablation) | RPNet-D8 · S3DIS (6-fold) | mIoU (%): 67.8 -> 69.1 | Improved |
| add semantic relation (summation) on top of geometric relations (ablation) | RPNet · S3DIS (6-fold) | mIoU (%): 64.2 -> 67.8 | Improved |
| add l1, difference and concatenated-coordinate geometric terms to l2 only (ablation) | RPNet · S3DIS (6-fold) | mIoU (%): 63.7 -> 64.2 | Improved |
| use summation or subtraction instead of concatenation or Hadamard product for semantic relation (ablation) | RPNet · S3DIS (6-fold) | mIoU (%): 67.8 vs 67.5 and 67.6 | Improved |
| use max-pooling instead of avg-pooling or sum-pooling for aggregation (ablation) | RPNet-D8 · S3DIS (6-fold) | mIoU (%): 67.5 -> 67.8 | Improved |
| increase depth of classification network from 3 to 7 (ablation) | RPNet-W1 · ModelNet40 | accuracy (%): 92.9 -> 92.1 | Hurt |
| increase width of segmentation network from 1 to 3 (ablation) | RPNet-D4 · Segmentation (dataset not stated) | accuracy (%): 66.8 -> 67.1 | No clear effect |
| increase classification width from RPNet-W9 to W15 (ablation) | RPNet-W · ModelNet40 | accuracy (%) | Hurt |
| increase segmentation depth of RPNet-D (ablation) | RPNet-D · ScanNet v2 | mIoU (%): 67.1 -> 68.2 | Improved |
| add normals to the input points | RPNet-W9 · ModelNet40 | accuracy (%): 93.9 -> 94.1 | Improved |
| use RPNet semantic relations instead of RS-CNN predefined relations for rigid-transform robustness | RPNet-W7 · ModelNet40 robustness | accuracy (%): 90.3 -> 90.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PointASNL — PointASNL: Robust Point Clouds Processing Using Nonlocal Neural Networks With Adaptive Sampling (CVPR 2020) · better on 4 of 4 shared results
- FPConv — FPConv: Learning Local Flattening for Point Convolution (CVPR 2020) · better on 3 of 3 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
- Structural Relational Reasoning of Point Clouds (CVPR 2019 · similarity 0.86)
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation (CVPR 2017 · similarity 0.86)
- Relationship Proposal Networks (CVPR 2017 · similarity 0.86)
- SGMNet: Learning Rotation-Invariant Point Cloud Representations via Sorted Gram Matrix (ICCV 2021 · similarity 0.86)
- Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025 · similarity 0.85)
- LogoSP: Local-global Grouping of Superpoints for Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2025 · similarity 0.85)
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.
