Research map / Detection and segmentation / 3D detection and driving
Efficient Transformer-based 3D Object Detection with Dynamic Token Halting
Mao Ye, Gregory P. Meyer, Yuning Chai, Qiang Liu
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 dynamic token halting (SST_halt) to SST_center | SST_center · Waymo val vehicle BEV AP [50, inf) L2 | BEV AP: 40.8 -> 41.2 | Improved |
| add improvements on top of halting (SST++halt) | SST_center · Waymo val vehicle BEV AP [50, inf) L2 | BEV AP: 40.8 -> 43.4 | Improved |
| add voxel (token) recycling for halted tokens (ablation) | SST++halt · Waymo val | Vehicle L1 AP: 61.9 -> 75.7 | Improved |
| add voxel recycling (ablation) | SST++halt · Waymo val | Cyclist L1 AP: 63.7 -> 70.0 | Improved |
| add voxel recycling (ablation) | SST++halt · Waymo val | Pedestrian L1 AP: 76.8 -> 78.7 | Improved |
| use non-uniform sparsity loss instead of uniform l1 (ablation) | SST++halt · Waymo val | Vehicle L1 AP: 74.5 -> 75.7 | Improved |
| use non-uniform sparsity loss instead of uniform l1 (ablation) | SST++halt · Waymo val | Cyclist L1 AP: 64.2 -> 70.0 | Improved |
| use non-uniform sparsity loss instead of uniform l1 (ablation) | SST++halt · Waymo val | Pedestrian L2 AP: 70.2 -> 72.3 | Improved |
| dynamic token halting vs width scaling at similar speedup (~1.1-1.2x) | SST_center · Waymo val | Vehicle L1 AP: 73.8 -> 76.1 | Improved |
| dynamic token halting vs adapted AViT at ~1.4x speedup | SST_center · Waymo val | Vehicle L1 AP: 70.3 -> 76.1 | Improved |
| token halting at 1.52x speedup (sparsity 0.89) vs original | SST_center · Waymo val | Vehicle L1 AP: 76.2 -> 75.4 | Mixed |
| apply SST++halt in place of SST_center on test split | SST_center · Waymo test | Vehicle L1 AP: 80.0 -> 81.2 | Improved |
Closest papers by meaning
- An End-to-End Transformer Model for 3D Object Detection (ICCV 2021 · similarity 0.87)
- Efficient Decoder-Free Object Detection with Transformers (ECCV 2022 · similarity 0.87)
- ViDT: An Efficient and Effective Fully Transformer-based Object Detector (ICLR 2022 · similarity 0.86)
- End-to-End Object Detection with Transformers (ECCV 2020 · similarity 0.86)
- Not All Tokens Matter All The Time: Dynamic Token Aggregation Towards Efficient Detection Transformers (ICML 2025 · similarity 0.85)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.85)
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.
