Research map / Detection and segmentation / 3D detection and driving
Revisiting Token Compression for Accelerating ViT-based Sparse Multi-View 3D Object Detectors
Mingqian Ji, Shanshan Zhang, Jian Yang
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 SPSS adaptive patch size (ablation) | StreamPETR ViT-L · nuScenes val | NDS / inference time: 61.2 -> 58.8 NDS; 317.0 -> 189.1 ms | Mixed |
| add CGFE after SPSS (ablation) | SPSS-enabled StreamPETR · nuScenes val | NDS: 58.8 -> 60.4 | Improved |
| add IPS after SPSS and CGFE (ablation) | SPSS+CGFE StreamPETR · nuScenes val | inference time: 209.6 -> 194.3 ms | Improved |
| use adaptive (18, 20) patch size instead of fixed 20 (ablation) | SEPatch3D · nuScenes val | NDS: 59.3 -> 60.3 | Improved |
| use adaptive (18, 20) patch size instead of fixed 18 (ablation) | SEPatch3D · nuScenes val | inference time: 242.6 -> 194.3 ms | Improved |
| use entropy-based selection instead of score-based (ablation) | IPS module · nuScenes val | NDS: 59.8 -> 60.3 | Improved |
| use entropy-based selection instead of attention-based (ablation) | IPS module · nuScenes val | NDS: 59.6 -> 60.3 | Improved |
| use adaptive selection instead of Top-K 30% (ablation) | IPS module · nuScenes val | NDS: 59.6 -> 60.3 | Improved |
| use spatial cue only instead of spatial and temporal cues (ablation) | SPSS · nuScenes val | NDS: 60.3 -> 59.7 | Hurt |
| use temporal cue only instead of spatial and temporal cues (ablation) | SPSS · nuScenes val | NDS: 60.3 -> 59.5 | Hurt |
| apply SEPatch3D instead of Cropr token pruning | StreamPETR · nuScenes val | mFP: 53.2% -> 40.5% | Improved |
| apply SEPatch3D-faster to DETR3D | DETR3D ViT-L · nuScenes val | inference time: 344.9 -> 216.2 ms | Improved |
Closest papers by meaning
- LiteVGGT: Boosting Vanilla VGGT via Geometry-aware Cached Token Merging (CVPR 2026 · similarity 0.88)
- Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More (ICML 2025 · similarity 0.88)
- Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers (NeurIPS 2026 · similarity 0.87)
- Patch-as-Decodable-Token: Towards Unified Multi-Modal Vision Tasks in MLLMs (ICLR 2026 · similarity 0.87)
- ApET: Approximation-Error Guided Token Compression for Efficient VLMs (CVPR 2026 · similarity 0.87)
- Merge3D: Efficient 3D Multimodal LLMs via Joint 2D-3D Token Merging (CVPR 2026 · 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.
