Research map / Point clouds / Point cloud classification
diffConv: Analyzing Irregular Point Clouds with an Irregular View
Manxi Lin, Aasa Feragen
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 masked attention (with balanced renormalization) to LS+DDBQ model (ablation) | diffConv · ModelNet40 | OA (%): 92.2 -> 93.6 | Improved |
| add density-dilated ball query instead of constant-radius ball query (ablation) | diffConv · ModelNet40 | OA (%): 93.1 -> 93.6 | Improved |
| add Laplacian smoothing (ablation) | diffConv · ModelNet40 | OA (%): 92.5 -> 93.6 | Improved |
| use balanced renormalization instead of original self-attention normalization (ablation) | diffConv · ModelNet40 | OA (%): 92.9 -> 93.6 | Improved |
| use density-dilated ball query instead of 20-NN grouping (ablation) | diffConv · ModelNet40-C | CER (%): 25.2 -> 21.4 | Improved |
| add masked attention instead of binary adjacency (ablation) | diffConv · ModelNet40-C | CER (%): 21.9 -> 21.4 | Improved |
| use masked attention instead of isotropic bias (ablation) | diffConv · ModelNet40 | OA (%): 92.2 -> 93.6 | Improved |
| use masked attention instead of spatial distance bias (ablation) | diffConv · ModelNet40 | OA (%): 90.2 -> 93.6 | Improved |
| use masked attention instead of inverse density bias (ablation) | diffConv · ModelNet40 | OA (%): 89.7 -> 93.6 | Improved |
| squared radius linear dilation instead of linear radius dilation (ablation) | diffConv · ModelNet40 | OA (%): 92.6 -> 93.6 | Improved |
| kernel density bandwidth 0.1 instead of 0.05 (ablation) | diffConv · ModelNet40 | OA (%): 93.1 -> 93.6 | Improved |
| increase squared initial radius from 0.005 to 0.1 (ablation) | diffConv · ModelNet40 | OA (%): 93.6 -> 91.8 | Hurt |
Closest papers by meaning
- AGConv: Adaptive Graph Convolution on 3D Point Clouds (TPAMI 2023 · similarity 0.90)
- Adaptive Graph Convolution for Point Cloud Analysis (ICCV 2021 · similarity 0.88)
- PointConv: Deep Convolutional Networks on 3D Point Clouds (CVPR 2019 · similarity 0.88)
- PointConvFormer: Revenge of the Point-Based Convolution (CVPR 2023 · similarity 0.88)
- Differentiable Convolution Search for Point Cloud Processing (ICCV 2021 · similarity 0.87)
- Convolution in the Cloud: Learning Deformable Kernels in 3D Graph Convolution Networks for Point Cloud Analysis (CVPR 2020 · similarity 0.87)
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.
