Research map / Point clouds / Point cloud classification
Efficient point sampling architectures
Improve point cloud backbones with sampling, scaling and spiking networks.
45 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Point cloud classification, keyword cluster “adversarial · convolution · invariant”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
14 papers in 2023–24 and 13 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.0% to 5.0%.
No longer reported by papers here: S3DIS Area-5 (mIoU, last in 2023).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · FoldingNet: Point Cloud Auto-Encoder via Deep Grid Deformation (CVPR 2018 · 1,382 citations)
- 2019 · Modeling Point Clouds With Self-Attention and Gumbel Subset Sampling (CVPR 2019 · 463 citations)
- 2020 · Grid-GCN for Fast and Scalable Point Cloud Learning (CVPR 2020 · 257 citations)
- 2021 · Regularization Strategy for Point Cloud via Rigidly Mixed Sample (CVPR 2021 · 64 citations)
- 2022 · PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (NeurIPS 2022 · 415 citations)
- 2023 · DSVT: Dynamic Sparse Voxel Transformer With Rotated Sets (CVPR 2023 · 179 citations)
- 2024 · LTA-PCS: Learnable Task-Agnostic Point Cloud Sampling (CVPR 2024 · 9 citations)
- 2025 · Spiking Point Transformer for Point Cloud Classification (AAAI 2025 · 13 citations)
- 2026 · Scalable Event Cloud Network for Event-based Classification (ICML 2026 · 1 citation)
Earliest papers here
- FoldingNet: Point Cloud Auto-Encoder via Deep Grid Deformation (CVPR 2018 · 1,382 citations)
- Training DNNs with Hybrid Block Floating Point (NeurIPS 2018 · 26 citations)
- Modeling Point Clouds With Self-Attention and Gumbel Subset Sampling (CVPR 2019 · 463 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
S3DIS Area-5 · mIoU (higher is better)
5 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 60.07 | Modeling Point Clouds With Self-Attention and Gumbel Subset Sampling (CVPR 2019 · 463 citations) |
| 2022 | 70.5 | PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (NeurIPS 2022 · 415 citations) |
ModelNet40 · OA (higher is better)
4 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 93.1 | Grid-GCN for Fast and Scalable Point Cloud Learning (CVPR 2020 · 257 citations) |
| 2022 | 94 | PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (NeurIPS 2022 · 415 citations) |
What has been tried
40 papers here report 483 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 8%, hurt 23%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Increase time steps from 1 to 4 | ModelNet40 test · OA | 90.87 -> 91.43 | Helped | Spiking Point Transformer for Point Cloud Classification (AAAI 2025 · 13 citations) |
| add both intra-view and inter-view semantic losses | ModelNet40, 128 points · instance accuracy | 87.8 -> 89.1 | Helped | LTA-PCS: Learnable Task-Agnostic Point Cloud Sampling (CVPR 2024 · 9 citations) |
| Replace the spiking E-3DSNN with its same-architecture ANN version | SemanticKITTI val · mIoU | 69.2 -> 69.4 | Helped | Efficient 3D Recognition with Event-driven Spike Sparse Convolution (AAAI 2025 · 8 citations) |
| use the hybrid architecture with ANN-based score estimation rather than the purely spiking architecture | PU-Net, 50K points, 2% noise · CD | 1.926 -> 1.798; reduction of 0.128 | Helped | Noise-Injected Spiking Graph Convolution for Energy-Efficient 3D Point Cloud Denoising (AAAI 2025 · 7 citations) |
| Use residual block design in Figure 2(b) instead of Figure 2(a) | ModelNet40 · OA | 89.4% vs. 90.6% | Helped | Point-to-Spike Residual Learning for Energy-Efficient 3D Point Cloud Classification (AAAI 2024 · 7 citations) |
| Add bucket rebalancing | nuScenes validation · mIoU | 78.6 -> 78.7 | Helped | Flash3D: Super-scaling Point Transformers through Joint Hardware-Geometry Locality (CVPR 2025 · 4 citations) |
| Add the HEPT kernel instead of the original transformer kernel and absolute positional encoding | Tracking-60k · AP@k | 72.28 -> 91.93 | Helped | Locality-Sensitive Hashing-Based Efficient Point Transformer with Applications in High-Energy Physics (ICML 2024 · 1 citation) |
| Fine-tune only GCD-sampling after loading pretrained PointNeXt weights | ScanObjectNN (PB T50 RS) · OA (%) | 87.40 -> 88.41 | Helped | GCD-Sampling: A General Cross-scale Decoupled Sampling for Point Cloud (AAAI 2025 · 0 citations) |
| Replace grid pseudo-coordinates with randomly sampled pseudo-coordinates | LSUN-Church-256 · FID-CLIP | 7.32 -> 11.66 | Hurt | INRFlow: Flow Matching for INRs in Ambient Space (ICML 2025 · 0 citations) |
| add semantic alignment (SA) via sorted feature matching | ModelNet10, ModelNet40, ShapeNet, and ScanObjectNN, PPC 3 · accuracy | 75.01 -> 84.42; 59.96 -> 72.08; 53.91 -> 63.74; 20.20 -> 31.84 | Helped | Dataset Distillation of 3D Point Clouds via Distribution Matching (NeurIPS 2025 · 0 citations) |
| Add PCA canonical-orientation projection | ModelNet40 Rotated · accuracy | 14.75 -> 60.77 | Helped | Point Cloud Dataset Distillation (ICML 2025) |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ModelNet40 · 5 of 9 papers
- ScanObjectNN · 4 of 9 papers
- NuScenes · 2 of 9 papers
- ShapeNetPart · 2 of 9 papers
- ModelNet10 · 2 of 9 papers
- ShapeNet · 2 of 9 papers
Baselines
- PointNet · 2 of 9 papers
- SpikingPointNet · 2 of 9 papers
- Herding · 2 of 9 papers
- K-Center · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 27 papers; from OpenAlex and the papers' affiliations).
- Anhui University · 2 papers
- Chinese Academy of Sciences · 2 papers
- Karlsruhe Institute of Technology · 2 papers
- Tsinghua University · 2 papers
- Georgia Institute of Technology · 2 papers
- Purdue University · 2 papers
Use it
Download BibTeX of 35 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Local feature aggregation networks (158 papers)
- Equivariant point cloud networks (84 papers)
- Adversarial point cloud attacks (37 papers)
Elsewhere in the topic
- Point-language and open-vocabulary 3D (107 papers)
- Scene flow and dynamic point clouds (89 papers)
- Self-supervised point cloud pretraining (73 papers)
- LiDAR point cloud compression and completion (60 papers)
- Point cloud detection and tracking (56 papers)
- Masked point autoencoders (22 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).