Research map / 3D vision / 3D generation
GSV2X: Geometry-Aware Uncertainty Modeling and Orthogonal Fusion for Robust Roadside Perception
Jianqiang Xu, Gensheng Pei, Huafeng Liu, Yazhou Yao
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 Gaussian-splatting BEV representation to BEVFusion-like baseline (ablation) | GSV2X-Base · RCooper intersection | AP@0.5: 43.6 -> 58.6 | Improved |
| add Orthogonal Fusion Module to Gaussian splatting model (ablation) | GSV2X-GS · RCooper intersection | AP@0.5: 58.6 -> 62.7 | Improved |
| add Geometry-guided Depth Estimation to Gaussian splatting model (ablation) | GSV2X-GS · RCooper intersection | AP@0.5: 58.6 -> 59.9 | Improved |
| add GDE and OFM together (full model) vs OFM only (ablation) | GSV2X · RCooper intersection | AP@0.7: 42.4 -> 40.3 | Hurt |
| add OFM only to Gaussian splatting model in corridor scene (ablation) | GSV2X-GS · RCooper corridor | AP@0.7: 40.5 -> 38.2 | Mixed |
| replace deterministic BEVFusion projection with Gaussian probabilistic lifting under perturbed calibration (ablation) | GSV2X-GS · RCooper intersection, perturbed calibration | AP@0.5 drop: -16.4 -> -3.2 | Improved |
| use learnable per-view perturbation factor instead of fixed lambda (ablation) | GSV2X · RCooper intersection | AP@0.5: 63.0 -> 63.4 | Improved |
| reduce active cameras from 4 to 1 at inference (ablation) | GSV2X · RCooper intersection | AP@0.5: 63.4 -> 55.6 | Hurt |
| GSV2X vs BEVFusion on pedestrian detection | GSV2X · DAIR-V2X-I val, Pedestrian Easy | AP3D|R40: 51.5 -> 58.3 | Improved |
Closest papers by meaning
- RESOLVE: A Multi-Resolution and Multi-Modal Dataset for Roadside Cooperative Perception (ECCV 2026 · similarity 0.88)
- RoScenes: A Large-scale Multi-view 3D Dataset for Roadside Perception (ECCV 2024 · similarity 0.88)
- Learnability-Driven Submodular Optimization for Active Roadside 3D Detection (CVPR 2026 · similarity 0.88)
- Effective Multi-sensor Conditioning for Street-view Novel-view Synthesis (NeurIPS 2026 · similarity 0.86)
- Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting (CVPR 2025 · similarity 0.85)
- RobuRCDet: Enhancing Robustness of Radar-Camera Fusion in Bird's Eye View for 3D Object Detection (ICLR 2025 · 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.
