Research map / Point clouds / Point cloud classification
RISurConv: Rotation Invariant Surface Attention-Augmented Convolutions for 3D Point Cloud Classification and Segmentation
Zhiyuan Zhang, Licheng Yang, Zhiyu Xiang
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 distance and triangle-relation features phi1-5 on top of normal angles (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 90.9 -> 96.0 | Improved |
| add normal-based tangent-space angle features (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 88.2 -> 96.0 | Improved |
| add distance L0 feature (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 95.5 -> 96.0 | Improved |
| add extra normal-angle features lambda and mu to RISP (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 96.0 -> 95.7 | No clear effect |
| add both self-attention layers and transformer encoder (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 92.8 -> 96.0 | Improved |
| add transformer encoder before FC layers (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 94.3 -> 96.0 | Improved |
| add SA1 in RISurConv (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 95.6 -> 96.0 | Improved |
| use 2 local triangle surfaces instead of 1 (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 94.7 -> 96.0 | Improved |
| use 3 or 4 local triangle surfaces instead of 2 (ablation) | RISurConv · ModelNet40 | overall accuracy (%): 96.0 -> 91.9 (3), 89.8 (4) | Hurt |
| rotation-invariant RISurConv vs non-invariant Point Transformer v2 under z/z | RISurConv · ModelNet40 (z/z) | overall accuracy (%): 94.2 -> 96.0 | Improved |
| remove transformer components to cut parameters (ablation) | RISurConv · ModelNet40 | parameters: 13.96M -> 1.35M | Improved |
Later papers that compared with this method
2 later accepted papers report RISurConv as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- 4D Local Modeling Toward Dynamic Global Perception for Ambiguity-free Rotation-Invariant Point Cloud Analysis (CVPR 2026 · better on 2 of 2 shared results)
- Enhancing Rotation-Invariant 3D Learning with Global Pose Awareness and Attention Mechanisms (AAAI 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RIConv++ — RIConv++: Effective Rotation Invariant Convolutions for 3D Point Clouds Deep Learning (IJCV 2022) · better on 4 of 4 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
- RIConv++: Effective Rotation Invariant Convolutions for 3D Point Clouds Deep Learning (IJCV 2022 · similarity 0.93)
- Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysis (ECCV 2020 · similarity 0.90)
- Pointwise Rotation-Invariant Network with Adaptive Sampling and 3D Spherical Voxel Convolution (AAAI 2020 · similarity 0.90)
- Rotation-Invariant Local-to-Global Representation Learning for 3D Point Cloud (NeurIPS 2020 · similarity 0.90)
- Rotation-Invariant Transformer for Point Cloud Matching (CVPR 2023 · similarity 0.89)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.88)
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.
