Research map / 3D vision / Depth and stereo
GVDepth: Zero-Shot Monocular Depth Estimation for Ground Vehicles based on Probabilistic Cue Fusion
Karlo Koledić, Luka Petrović, Ivan Marković, Ivan Petrović
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 vertical canonical transform (ground-plane cue) instead of standard depth regression (ablation) | GVDepth (KITTI-trained) · KITTI-trained, tested on DrivingStereo | A.Rel (%): 25.60 -> 10.43 | Improved |
| add probabilistic fusion of focal and vertical cues (ablation) | GVDepth (KITTI-trained) · KITTI-trained, tested on DDAD | A.Rel (%): 14.26 -> 11.81 | Improved |
| use vertical canonical transform rather than focal canonical transform (ablation) | GVDepth (Waymo-trained) · Waymo-trained, tested on KITTI | A.Rel (%): 18.21 -> 8.30 | Improved |
| use vertical canonical transform rather than focal on DDAD-trained model tested on KITTI (ablation) | GVDepth (DDAD-trained) · DDAD-trained, tested on KITTI | A.Rel (%): 12.45 -> 22.21 | Hurt |
| fuse cues instead of vertical only when object size cue is stronger (ablation) | GVDepth (KITTI-trained) · KITTI-trained, tested on Argoverse | A.Rel (%): 14.72 -> 10.45 | Improved |
| use fusion instead of vertical only on Waymo-trained model tested on KITTI (ablation) | GVDepth (Waymo-trained) · Waymo-trained, tested on KITTI | A.Rel (%): 8.30 -> 10.92 | Hurt |
| replace GEDepth ground-attention formulation with vertical canonical transform (ablation) | GVDepth (KITTI-trained) · KITTI-trained, tested on Waymo | A.Rel (%): 22.11 -> 13.42 | Improved |
| add Waymo side-camera images to training (ablation) | Vertical model · Waymo to DDAD front-left | A.Rel (%): 19.21 -> 16.44 | Improved |
| use vertical transform instead of focal on back camera (ablation) | Waymo-trained models · Waymo to DDAD back camera | A.Rel (%): 20.93 -> 22.12 | Hurt |
| train GVDepth on a single dataset versus multi-dataset zero-shot methods | GVDepth · KITTI (zero-shot) | A.Rel (%): 6.96 vs 7.91 (Metric3D) | Mixed |
Closest papers by meaning
- Towards Zero-Shot Scale-Aware Monocular Depth Estimation (ICCV 2023 · similarity 0.89)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.88)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.88)
- Adaptive Fusion of Single-View and Multi-View Depth for Autonomous Driving (CVPR 2024 · similarity 0.87)
- UniDepth: Universal Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.87)
- BetterDepth: Plug-and-Play Diffusion Refiner for Zero-Shot Monocular Depth Estimation (NeurIPS 2024 · similarity 0.87)
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.
