Research map / 3D vision / 3D generation
CAPE: Camera View Position Embedding for Multi-View 3D Object Detection
Kaixin Xiong, Shi Gong, Xiaoqing Ye, Xiao Tan, Ji Wan, Errui Ding, Jingdong Wang, Xiang Bai
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 camera-view position embeddings with bilateral attention instead of global position embeddings with bilateral attention (ablation) | CAPE · nuScenes validation | NDS: +1.4% | Improved |
| Add bilateral attention when using camera-view position embeddings (ablation) | CAPE · nuScenes validation | NDS: 0.025 -> 0.479 | Improved |
| Add query feature-guided position embedding (ablation) | CAPE with K-FPE · nuScenes validation | NDS: +1.6% | Improved |
| Add key feature-guided position embedding (ablation) | CAPE without Q-FPE · nuScenes validation | NDS: +1.6% | Improved |
| Add query feature-guided position embedding (ablation) | CAPE without K-FPE · nuScenes validation | mAOE: 0.515 -> 0.551 | Hurt |
| Use a separate query set for each frame instead of sharing queries (ablation) | CAPE-T · nuScenes validation | NDS: +0.9% | Improved |
| Add auxiliary supervision on previous frames (ablation) | CAPE-T with separate frame queries · nuScenes validation | mAP: +0.5% | Improved |
| Add ego-motion embedding to concat + MLP fusion (ablation) | CAPE-T temporal fusion · nuScenes validation | NDS: 0.527 -> 0.531 | Improved |
| Use channel attention instead of concat + MLP for temporal fusion (ablation) | CAPE-T without ego-motion embedding · nuScenes validation | NDS: 0.527 -> 0.530 | Improved |
| Add camera-view position embeddings with bilateral attention | CAPE versus PETR · nuScenes test | NDS: 0.504 -> 0.520 | Improved |
| Extend CAPE with temporal modeling | CAPE · nuScenes test | NDS: 0.520 -> 0.610 | Improved |
| Use camera-view local position embeddings | CAPE-T versus PETRv2 · nuScenes validation with extrinsic rotation noise, R_max = 4 | mAP drop: 1.31% vs 2.39% | Improved |
Closest papers by meaning
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.86)
- Towards Generalizable Multi-Camera 3D Object Detection via Perspective Rendering (AAAI 2025 · similarity 0.85)
- PromptDet: A Lightweight 3D Object Detection Framework with LiDAR Prompts (AAAI 2025 · similarity 0.85)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.85)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.85)
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 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.
