Research map / Point clouds / Point cloud classification
Cloud Transformers: A Universal Approach to Point Cloud Processing Tasks
Kirill Mazur, Victor Lempitsky
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 |
|---|---|---|---|
| use element-wise max instead of sum aggregation in rasterization (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 57.9 -> 63.7 | Improved |
| use element-wise max instead of mean aggregation (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 61.3 -> 63.7 | Improved |
| use element-wise max instead of sum aggregation (ablation) | Cloud Transformer · ScanObjectNN | accuracy: 82.1 -> 85.0 | Improved |
| predict keys with a learned deep residual instead of linear layer (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 62.5 -> 63.7 | Improved |
| predict keys with a learned deep residual instead of linear layer (ablation) | Cloud Transformer · ScanObjectNN | accuracy: 81.9 -> 85.0 | Improved |
| learn the key projection instead of fixed random rotation (ablation) | Cloud Transformer · ScanObjectNN | accuracy: 84.9 -> 85.0 | No clear effect |
| learn the key projection instead of fixed random rotation (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 62.5 -> 63.7 | Improved |
| add planar heads alongside volumetric heads (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 63.4 -> 63.7 | Improved |
| use 16 heads instead of a single wide head (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 63.1 -> 63.7 | Improved |
| use full-resolution rather than coarser feature maps (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 62.2 -> 63.7 | Improved |
| use four CMHCT blocks instead of two (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 62.2 -> 63.7 | Improved |
| use eight CMHCT blocks instead of four (ablation) | Cloud Transformer · S3DIS Area-5 | mIoU: 63.7 -> 64.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- GRNet — GRNet: Gridding Residual Network for Dense Point Cloud Completion (ECCV 2020) · better on 1 of 2 shared results
- PointCNN — PointCNN: Convolution On X-Transformed Points (NeurIPS 2018) · better on 2 of 2 shared results
- JSENet† — JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds (ECCV 2020) · better on 1 of 1 shared result
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
- LitePT: Lighter Yet Stronger Point Transformer (CVPR 2026 · similarity 0.91)
- Spatial Transformer for 3D Point Clouds (TPAMI 2021 · similarity 0.91)
- Fully-Convolutional Point Networks for Large-Scale Point Clouds (ECCV 2018 · similarity 0.90)
- A Network Architecture for Point Cloud Classification via Automatic Depth Images Generation (CVPR 2018 · similarity 0.90)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.89)
- Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · 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.
