Research map / Detection and segmentation / 3D detection and driving
Object as Query: Lifting Any 2D Object Detector to 3D Detection
Zitian Wang, Zehao Huang, Jiahui Fu, Naiyan Wang, Si Liu
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 |
|---|---|---|---|
| replace fixed queries with dynamically generated queries from a 2D detector (ablation) | MV2D (ResNet-50) · nuScenes val | mAP: 36.2 -> 38.0 | Improved |
| increase number of fixed queries from 300 to 1500 (ablation) | fixed-query baseline · nuScenes val | NDS: 39.4 -> 40.6 | Improved |
| restrict cross attention to relevant features by top1 IoU rule (ablation) | MV2D · nuScenes val | NDS: 41.2 -> 42.4 | Improved |
| use all overlapped rule instead of top1 IoU (ablation) | MV2D · nuScenes val | NDS: 42.4 -> 42.6 | Improved |
| add history frame input (ablation) | MV2D · nuScenes val | NDS: 42.6 -> 51.1 | Improved |
| use MV2D instead of fixed-query baseline with history (ablation) | MV2D · nuScenes val | NDS: 47.4 -> 51.1 | Improved |
| use query generator with equivalent intrinsics instead of uniform depth sampling or scale-based depth (ablation) | MV2D · nuScenes val | NDS: 40.1 / 40.8 -> 42.6 | Improved |
| use relevant-object-region cross attention instead of deformable attention (ablation) | MV2D · nuScenes val | NDS: 48.4 -> 51.1 | Improved |
| use lighter 2D detector (YOLOX) instead of Faster R-CNN (ablation) | MV2D · nuScenes val | mAP: 41.4 -> 41.3 | No clear effect |
| use nuImages pre-training instead of COCO (ablation) | MV2D · nuScenes val | NDS: 49.8 -> 51.1 | Improved |
| increase decoder layers from 1 to 2 (ablation) | MV2D · nuScenes val | mAP: 36.4 -> 39.1 (+2.7) | Improved |
| use MV2D-T instead of PETRv2 on val set | ResNet-101 · nuScenes val | mAP: 0.421 -> 0.471 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- PETRv2 — PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images (ICCV 2023) · better on 6 of 6 shared results
- PolarFormer — PolarFormer: Multi-Camera 3D Object Detection with Polar Transformer (AAAI 2023) · better on 4 of 4 shared results
- PETR — PETR: Position Embedding Transformation for Multi-View 3D Object Detection (ECCV 2022) · better on 2 of 2 shared results
- BEVFormer — BEVFormer: Learning Bird’s-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers (ECCV 2022) · better on 1 of 1 shared result
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
- Enhancing 3D Object Detection with 2D Detection-Guided Query Anchors (CVPR 2024 · similarity 0.90)
- 2D-Driven 3D Object Detection in RGB-D Images (ICCV 2017 · similarity 0.90)
- Multi-View Attentive Contextualization for Multi-View 3D Object Detection (CVPR 2024 · similarity 0.88)
- MVSDet: Multi-View Indoor 3D Object Detection via Efficient Plane Sweeps (NeurIPS 2024 · similarity 0.87)
- Far3D: Expanding the Horizon for Surround-View 3D Object Detection (AAAI 2024 · similarity 0.86)
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion (CVPR 2025 · 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.
