ExQuery: Explicit Structured Query Priors for Temporal 3D Object Detection
Organizations: Graduate School of Information, Production and Systems, Waseda University, Fukuoka, Japan
Abstract
Sparse query-based 3D detectors commonly encode heterogeneous semantic, geometric, and motion information into unified latent queries, including when explicit 3D anchors are used. We introduce ExQuery, an explicit structured query framework for temporal 3D object detection. Each decoded query is organized into a shared semantic context slice and physical-state-specific slices for center, dimensions, orientation, and velocity. This structure is preserved from prior encoding to branch-specific regression and is reused to reconstruct temporal queries: predicted physical states are transformed by ego motion, classification scores form soft semantic priors, and the reconstructed query is fused with the stored latent instance feature through gated affine modulation. Experiments on the nuScenes benchmark show that our method achieves competitive detection performance, and ablation studies validate the effectiveness of our proposed modules.
Figures & tables
| Method | Backbone | Resolution | NDS | mAP | mATE | mASE | mAOE | mAVE | mAAE |
| PETRv2 [ Liu et al.(2023b)Liu, Yan, Jia, Li, Gao, Wang, and Zhang ] | ResNet50 | 256 704 | 0.456 | 0.349 | 0.700 | 0.275 | 0.580 | 0.437 | 0.187 |
| BEVDet4D [ Huang and Huang(2022) ] | ResNet50 | 256 704 | 0.457 | 0.322 | 0.703 | 0.278 | 0.495 | 0.354 | 0.206 |
| BEVDepth [ Li et al.(2023b)Li, Ge, Yu, Yang, Wang, Shi, Sun, and Li ] | ResNet50 | 256 704 | 0.475 | 0.351 | 0.639 | 0.267 | 0.479 | 0.428 | 0.198 |
| BEVStereo [ Li et al.(2023a)Li, Bao, Ge, Yang, Sun, and Li ] | ResNet50 | 256 704 | 0.500 | 0.372 | 0.598 | 0.270 | 0.438 | 0.367 | 0.190 |
| SOLOFusion [ Park et al.(2022)Park, Xu, Yang, Keutzer, Kitani, Tomizuka, and Zhan ] | ResNet50 | 256 704 | 0.534 | 0.427 | 0.567 | 0.274 | 0.511 | 0.252 | 0.181 |
| Sparse4Dv2 [ Lin et al.(2023a)Lin, Lin, Pei, Huang, and Su ] | ResNet50 | 256 704 | 0.539 | 0.439 | 0.598 | 0.270 | 0.475 | 0.282 | 0.179 |
| Model | Params. | GFLOPs | Mem. | FPS | Latency |
| (M) | (MB) | (ms) | |||
| RepDETR3D | 43.29 | 225.03 | 404.12 | 67.22 | 14.88 |
| \method | 46.17 | 227.60 | 415.69 | 54.10 | 18.48 |
| Structured Query | Explicit Prior | Gated Fusion | NDS | mAP |
| 0.470 | 0.383 | |||
| ✓ | 0.497 | 0.386 | ||
| ✓ | 0.501 | 0.391 | ||
| ✓ | ✓ | 0.516 | 0.406 | |
| ✓ | ✓ | ✓ | 0.522 | 0.417 |
3D Object Detection
3D object detection identifies objects and estimates their locations, sizes, and orientations from images, point clouds, radar, or fused sensors.
See all 195 papers