Research map / Detection and segmentation / 3D detection and driving
Parametric Depth Based Feature Representation Learning for Object Detection and Segmentation in Bird's-Eye View
Jiayu Yang, Enze Xie, Miaomiao Liu, Jose M. Alvarez
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 parametric depth lifting together with occupancy-aware aggregation (ablation) | BEV feature transformation · nuScenes val (ResNet-50) | mAP: 0.408 -> 0.424 (uniform + PointPillars to full) | Improved |
| replace PointPillars aggregation with occupancy-aware aggregation (parametric lifting) (ablation) | parametric depth lifting · nuScenes val (ResNet-50) | mAP: 0.410 -> 0.424 | Improved |
| replace PointPillars aggregation with occupancy-aware aggregation (non-parametric or uniform lifting) (ablation) | LSS or uniform lifting · nuScenes val (ResNet-50) | mAP: 0.409 -> 0.414 (non-parametric); 0.408 -> 0.413 (uniform) | Improved |
| replace parametric depth lifting with parametric lifting only with PointPillars aggregation (ablation) | BEV feature lifting · nuScenes val (ResNet-50) | mAP: 0.408 -> 0.410 | No clear effect |
| remove LiDAR depth supervision from parametric depth (ablation) | Ours · nuScenes val BEV segmentation | Road IoU: 78.7 -> 77.9 | Hurt |
| use parametric depth method versus M2BEV on visible regions | BEV segmentation · nuScenes val visible region | Road IoU: 82.9 -> 84.8 | Improved |
| use parametric depth method versus M2BEV on occluded regions | BEV segmentation · nuScenes val occluded region | Lane IoU: 12.8 -> 12.4 | No clear effect |
| use parametric depth instead of M2BEV uniform depth in joint multi-task setting | M2BEV · nuScenes val joint | mAP / NDS / Road / Lane: mAP 0.408 -> 0.424; NDS 0.454 -> 0.467 | Improved |
| predict extra depth parameters (parametric depth) versus uniform depth | BEV model · nuScenes efficiency | memory / FPS: memory 7718 -> 8902; FPS 1.2 -> 1.3 | Mixed |
Closest papers by meaning
- Instance-Aware Multi-Camera 3D Object Detection with Structural Priors Mining and Self-Boosting Learning (AAAI 2024 · similarity 0.88)
- Improving Bird's Eye View Semantic Segmentation by Task Decomposition (CVPR 2024 · similarity 0.88)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.87)
- Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting (CVPR 2025 · similarity 0.86)
- SDGOCC: Semantic and Depth-Guided Bird's-Eye View Transformation for 3D Multimodal Occupancy Prediction (CVPR 2025 · similarity 0.86)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (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.
