Research map / Detection and segmentation / 3D detection and driving
MsSVT++: Mixed-Scale Sparse Voxel Transformer With Center Voting for 3D Object Detection
Jianan Li, Shaocong Dong, Lihe Ding, Tingfa Xu
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 Balanced Multi-window Sampling replacing dilated key sampling (ablation) | 3D window-attention baseline · Waymo Pedestrian (20% train) | mAP: 71.66 -> 73.44 | Improved |
| add Balanced Multi-window Sampling replacing dilated key sampling (ablation) | 3D window-attention baseline · Waymo Cyclist (20% train) | mAP: 63.31 -> 66.88 | Improved |
| add Scale-aware Head Attention (ablation) | BMS model · Waymo Pedestrian (20% train) | mAP: 73.44 -> 75.08 | Improved |
| add Scale-aware Relative Position Encoding (ablation) | BMS+SHA model · Waymo Vehicle (20% train) | mAP: 71.96 -> 72.37 | Improved |
| add Center Voting with averaged point features (ablation) | BMS+SHA+SRPE model · Waymo Vehicle (20% train) | mAP: 72.37 -> 73.63 | Improved |
| add mixed-scale contextual features to voted voxels (ablation) | Center Voting module · Waymo Vehicle (20% train) | mAP: 73.63 -> 73.86 | Improved |
| add mixed-scale contextual features to voted voxels (ablation) | Center Voting module · Waymo Pedestrian (20% train) | mAP: 76.53 -> 76.47 | No clear effect |
| add Center Voting module (full pipeline) overall (ablation) | MsSVT++ · Waymo Vehicle/Cyclist | mAP: +1.49 vehicle, +1.28 cyclist, +0.48 pedestrian | Improved |
| use single-scale key window instead of mixed scales (ablation) | MsSVT++ key windows · Waymo (ablation) | mAP: 71.46 / 74.39 / 65.55 for [3,3,5] only | Hurt |
| add more key window scales ([5,5,7], [9,9,7]) (ablation) | MsSVT++ key windows · Waymo Cyclist (ablation) | mAP: 67.01 -> 66.53 | Hurt |
| increase key window size (ablation) | MsSVT++ key windows · Waymo (ablation) | mAP | Hurt |
| use MsSVT++ SS instead of SST on 100% data | SST-SS-SF · Waymo Vehicle (100% train) | mAP: 75.13 -> 78.96 | Improved |
Closest papers by meaning
- MsSVT: Mixed-scale Sparse Voxel Transformer for 3D Object Detection on Point Clouds (NeurIPS 2022 · similarity 0.95)
- Voxel Transformer for 3D Object Detection (ICCV 2021 · similarity 0.91)
- VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (CVPR 2023 · similarity 0.90)
- PVT-SSD: Single-Stage 3D Object Detector With Point-Voxel Transformer (CVPR 2023 · similarity 0.90)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.89)
- Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection (NeurIPS 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.
