Research map / Detection and segmentation / 3D detection and driving
3D Small Object Detection with Dynamic Spatial Pruning
Zhihao Sun, Ziwei Wang, Hongmin Liu, Jie Zhou, Jiwen Lu, Xiuwei 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 |
|---|---|---|---|
| double the spatial resolution of TR3D | TR3D · 3D small object detection | mAP_S: 52.7% -> 62.8% | Improved |
| double the spatial resolution of TR3D | TR3D · ScanNet-md40 | memory: 1250 -> 4450 | Hurt |
| add dynamic spatial pruning (tau=0.3) to the higher-resolution TR3D | TR3D-higher · ScanNet-md40 | speed (FPS): 5.2 -> 12.5 | Improved |
| add dynamic spatial pruning (tau=0.3) | TR3D-higher · ScanNet-md40 | memory: 4450 -> 700 | Improved |
| increase pruning threshold from 0 to 0.3 (ablation) | DSPDet3D · ScanNet-md40 | mAP_S@0.25: 44.79 -> 43.77 | Hurt |
| use DSPDet3D (tau=0) instead of TR3D-higher | TR3D-higher · ScanNet-md40 | mAP_S@0.25: 41.70 -> 44.79 | Improved |
| use partial addition instead of removing it (ablation) | DSP module · ScanNet-md40 (10 FPS) | mAP: 55.3 -> 65.1 | Improved |
| use partial addition instead of union addition (ablation) | DSP module · ScanNet-md40 (10 FPS) | mAP: 57.9 -> 65.1 | Improved |
| use affecting-field-shaped keeping mask instead of spherical mask (ablation) | DSP module · ScanNet-md40 (10 FPS) | mAP: 63.0 -> 65.1 | Improved |
| sample positives inside P-cube instead of inside bounding box (ablation) | DSP module · ScanNet-md40 (10 FPS) | mAP: 62.4 -> 65.1 | Improved |
| add training-time weak pruning (ablation) | DSP module · ScanNet-md40 | trainability | Improved |
| set r=7 (with P=7) instead of smaller r (ablation) | DSPDet3D · ScanNet-md40 | mAP | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CAGroup3D — CAGroup3D: Class-Aware Grouping for 3D Object Detection on Point Clouds (NeurIPS 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Spatial Pruned Sparse Convolution for Efficient 3D Object Detection (NeurIPS 2022 · similarity 0.88)
- Accelerate 3D Object Detection Models via Zero-Shot Attention Key Pruning (ICCV 2025 · similarity 0.87)
- RBGNet: Ray-Based Grouping for 3D Object Detection (CVPR 2022 · similarity 0.87)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.86)
- UNION: Unsupervised 3D Object Detection using Object Appearance-based Pseudo-Classes (NeurIPS 2024 · similarity 0.86)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · 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.
