Research map / Point clouds / Point cloud classification
ScatterFormer: Efficient Voxel Transformer with Scattered Linear Attention
Chenhang He, Ruihuang Li, Guowen Zhang, Lei Zhang
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 Scattered Linear Attention module (ablation) | ScatterFormer · Waymo 20% ablation | L2 APH (Veh/Ped/Cyc): 68.0 -> 71.1 (Veh) | Improved |
| add Cross-Window Interaction module (ablation) | ScatterFormer · Waymo 20% ablation | L2 APH (Veh/Ped/Cyc): 69.2 -> 71.1 (Veh) | Improved |
| use CWI instead of shifted window (ablation) | ScatterFormer · Waymo 20% ablation | Vehicle L2 APH: 69.5 -> 71.1 | Improved |
| add conditional position encoding (ablation) | ScatterFormer · Waymo 20% ablation | Vehicle L2 APH: 69.8 -> 71.1 | Improved |
| increase window size beyond 12 (ablation) | ScatterFormer · Waymo 20% (window size) | Pedestrian L2 APH: 70.9 -> 69.2 (size 16) | Hurt |
| use 1D decomposed kernels instead of 5x5 2D kernel in CWI (ablation) | CWI module · Waymo 20% (CWI kernel) | Vehicle L2 APH: 69.5 -> 71.1 | Improved |
| use 7x7 2D kernel in CWI instead of 1D kernels (ablation) | CWI module · Waymo 20% (CWI kernel) | latency: 47ms -> 65ms | Mixed |
| swap Efficient Attention for other linear attentions (ablation) | SLA module · Waymo 20% (linear attention) | L2 APH: 71.1 -> 69.9 (Gated), 71.5 (XCA) Veh | No clear effect |
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.
- HEDNet — HEDNet: A Hierarchical Encoder-Decoder Network for 3D Object Detection in Point Clouds (NeurIPS 2023) · 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
- Voxel Transformer for 3D Object Detection (ICCV 2021 · similarity 0.87)
- FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformer (CVPR 2023 · similarity 0.87)
- SCube: Instant Large-Scale Scene Reconstruction using VoxSplats (NeurIPS 2024 · similarity 0.85)
- Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · similarity 0.85)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.84)
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.84)
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.
