Research map / Point clouds / Point cloud classification
DSVT: Dynamic Sparse Voxel Transformer With Rotated Sets
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DSVT-V instead of the strongest reported one-stage baseline | Waymo Open single-frame detection · Waymo Open val, L2 | mAPH: +3.2 | Helped |
| Use DSVT-P instead of ResBackbone1x | Waymo Open detection backbone · Waymo val, L2 | mAPH: +1.78 | Helped |
| Use rotated set partitioning, alternating X- and Y-axis ordering (ablation) | DSVT sparse attention · Waymo val, L2 | mAPH: 61.03 -> 61.40 | Helped |
| Use regional partitioning instead of random voxel sampling (ablation) | DSVT sparse attention · Waymo val, L2 | mAPH: +0.83 | Helped |
| Use attention-style 3D pooling instead of linear pooling (ablation) | DSVT 3D pooling · Waymo val, L2 | mAPH: 68.60 -> 69.31 | Helped |
| Add key-padding masks for empty voxels to attention-style pooling (ablation) | DSVT 3D pooling · Waymo val, L2 | mAPH: 69.31 -> 68.94 | Hurt |
| Use hybrid window sizes (12, 12) and (24, 24) instead of equal (12, 12) windows (ablation) | DSVT window attention · Waymo val, L2 | mAPH: 68.35 -> 68.59 | Helped |
| Increase the second window size from (24, 24) to (36, 36) (ablation) | DSVT window attention · Waymo val, L2 | mAPH: 68.59 -> 68.18 | Hurt |
| Increase DSVT-P set size from 24 to 36 (ablation) | DSVT-P · Waymo val | mAPH: 68.14 -> 68.59 | Helped |
| Increase DSVT-P set size from 36 to 48 (ablation) | DSVT-P · Waymo val | mAPH: 68.59 -> 68.43 | Hurt |
| Increase DSVT-V set size from 36 to 48 (ablation) | DSVT-V · Waymo val | mAPH: 69.31 -> 69.67 | Helped |
| Increase DSVT-V set size from 48 to 60 (ablation) | DSVT-V · Waymo val | mAPH: 69.67 -> 69.60 | Hurt |
Closest papers by meaning
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.88)
- CVT-xRF: Contrastive In-Voxel Transformer for 3D Consistent Radiance Fields from Sparse Inputs (CVPR 2024 · similarity 0.86)
- GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction (NeurIPS 2025 · similarity 0.86)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · similarity 0.86)
- TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention (NeurIPS 2026 · similarity 0.86)
- MeshFormer : High-Quality Mesh Generation with 3D-Guided Reconstruction Model (NeurIPS 2024 · similarity 0.86)
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.