Research map / Detection and segmentation / 3D detection and driving
Accelerate 3D Object Detection Models via Zero-Shot Attention Key Pruning
Lizhen Xu, Xiuxiu Bai, Xiaojun Jia, Jianwu Fang, Shanmin Pang
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 |
|---|---|---|---|
| use max instead of mean of classification vector for key importance (ablation) | tgGBC on StreamPETR-VovNet · nuScenes val | mAP (%): 44.12% -> 48.55% | Improved |
| use max instead of min of classification vector for key importance (ablation) | tgGBC on StreamPETR-VovNet · nuScenes val | mAP (%): 34.88% -> 48.55% | Improved |
| use max instead of mean of classification vector for key importance (ablation) | tgGBC on OPEN-R101 · nuScenes val | mAP (%): 49.57% -> 51.57% | Improved |
| use max instead of min of classification vector for key importance (ablation) | tgGBC on OPEN-R101 · nuScenes val | mAP (%): 47.65% -> 51.57% | Improved |
| weight attention by classification scores in key importance (ablation) | tgGBC · nuScenes val | mAP and NDS | Improved |
| use n=2 tgGBC layers instead of n=1 (ablation) | tgGBC on StreamPETR-VovNet · nuScenes val | mAP (%): 47.84% -> 48.55% | Improved |
| use n=2 tgGBC layers instead of n=1 (ablation) | tgGBC on StreamPETR-VovNet · nuScenes val | decoder time (ms): 31.61 -> 34.98 | Hurt |
| use n=5 instead of n=3 tgGBC layers (ablation) | tgGBC on StreamPETR-VovNet · nuScenes val | mAP (%): 48.21% -> 48.92% | Mixed |
| increase number of pruned keys r (ablation) | tgGBC · nuScenes val | mAP and NDS | Mixed |
| replace ToMe with tgGBC | StreamPETR-R50-704x256 · nuScenes val | mAP (%): 36.12% -> 37.93% | Improved |
| replace ToMe with tgGBC | OPEN-VovNet-800x320 · nuScenes val | mAP (%): 49.21% -> 52.12% | Improved |
| add tgGBC pruning to ToC3D | ToC3D · nuScenes val | decoder time (ms): 77.92 -> 39.17 | Improved |
Closest papers by meaning
- 3D Small Object Detection with Dynamic Spatial Pruning (ECCV 2024 · similarity 0.87)
- An End-to-End Transformer Model for 3D Object Detection (ICCV 2021 · similarity 0.86)
- Redundant Queries in DETR-Based 3D Detection Methods: Unnecessary and Prunable (AAAI 2026 · similarity 0.86)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.86)
- Zero-TPrune: Zero-Shot Token Pruning through Leveraging of the Attention Graph in Pre-Trained Transformers (CVPR 2024 · similarity 0.85)
- SP3D: Boosting Sparsely-Supervised 3D Object Detection via Accurate Cross-Modal Semantic Prompts (CVPR 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.
