Research map / Point clouds / Segmentation and detection
Ada3D : Exploiting the Spatial Redundancy with Adaptive Inference for Efficient 3D Object Detection
Tianchen Zhao, Xuefei Ning, Ke Hong, Zhongyuan Qiu, Pu Lu, Yali Zhao, Linfeng Zhang, Lipu Zhou, Guohao Dai, Huazhong Yang, Yu Wang
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 density-guided spatial filtering to the importance predictor (ablation) | Ada3D-B · KITTI val | mAP (Mod.): 65.1 -> 66.1 | Improved |
| add density guidance at 50% drop ratio (ablation) | Ada3D importance predictor · KITTI val | Ped. Mod. AP: 50.2 -> 53.5 | Improved |
| add learned importance predictor to density-only criterion (ablation) | Ada3D · KITTI val (R_drop 50%) | Ped. Mod. AP: 39.4 -> 53.5 | Improved |
| replace normal BN with sparsity-preserving BN in the 2D backbone (ablation) | CenterPoint · KITTI val | mAP (Mod.): 66.1 -> 66.0 | No clear effect |
| use SP-BN instead of Nonzero BN (ablation) | CenterPoint 2D backbone · KITTI val | Car Mod. AP: 74.5 -> 79.1 | Improved |
| adaptive inference on 2D backbone only (Ada3D-A) (ablation) | CenterPoint · KITTI val | mAP (Mod.): 66.1 -> 66.4 | Improved |
| increase drop rate to Ada3D-C (60% voxels, 90% pixels) (ablation) | CenterPoint · KITTI val | mAP (Mod.): 66.1 -> 65.4 | Hurt |
| apply Ada3D versus model width halving / output pruning | CenterPoint nuScenes · nuScenes val (voxel=0.075) | mAP: 58.62 vs 57.19 (0.5W) vs 57.80 (SPSS-Conv); 3.34x vs 2.78x vs 1.14x FLOPs | Improved |
| combine Ada3D with SPVNAS-searched backbone | CenterPoint SPVNAS · KITTI val | FLOPs optimization: 1.07x -> 3.95x FLOPs; mAP 65.5 -> 65.5 | Improved |
| apply Ada3D to fully sparse VoxelNeXt | VoxelNeXt · nuScenes val | mAP: 60.50 -> 59.75 mAP; 1.19x FLOPs | Mixed |
| apply Ada3D on nuScenes (low-resolution BEV) for latency | CenterPoint nuScenes · nuScenes val | GPU latency | No clear effect |
Closest papers by meaning
- Point Density-Aware Voxels for LiDAR 3D Object Detection (CVPR 2022 · similarity 0.90)
- MsSVT: Mixed-scale Sparse Voxel Transformer for 3D Object Detection on Point Clouds (NeurIPS 2022 · similarity 0.89)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.89)
- SwiftPillars: High-Efficiency Pillar Encoder for Lidar-Based 3D Detection (AAAI 2024 · similarity 0.88)
- BEVSpread: Spread Voxel Pooling for Bird's-Eye-View Representation in Vision-based Roadside 3D Object Detection (CVPR 2024 · similarity 0.88)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 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.
