Research map / 3D vision / Depth and stereo
Camera Height Doesn't Change: Unsupervised Training for Metric Monocular Road-Scene Depth Estimation
Genki Kinoshita, Ko Nishino
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 FUMET training to Monodepth2 | Monodepth2 · KITTI Eigen (median scaling) | AbsRel: 0.115 -> 0.106 | Improved |
| optimize camera height across frames and epochs rather than per-frame size prior (ablation) | FUMET · KITTI | AbsRel: 0.115 -> 0.108 | Improved |
| add outlier filtering in Silhouette Projector (ablation) | FUMET · KITTI | AbsRel: 0.116 -> 0.108 | Improved |
| balance loss weights when only one of aux or cam loss is used (ablation) | FUMET · KITTI | AbsRel: 0.125 -> 0.149 (aux only); 0.116 -> 0.211 (cam only) | Hurt |
| balance loss weights when both aux and cam loss are used (ablation) | FUMET · KITTI | AbsRel: 0.122 -> 0.108 | Improved |
| train FUMET on mixed datasets plus KITTI instead of KITTI only | FUMET · KITTI Eigen (832x512) | AbsRel: 0.103 -> 0.082 | Improved |
| fix camera height computed offline from pre-trained model instead of online optimization (ablation) | FUMET · KITTI | AbsRel: 0.108 -> 0.106 | Improved |
| add YouTube in-the-wild videos to mixed plus KITTI training | FUMET · KITTI Eigen (832x512) | AbsRel: 0.082 -> 0.084 | No clear effect |
| use recovered COLMAP intrinsics instead of ground-truth intrinsics | FUMET · KITTI | AbsRel: 0.108 -> 0.110 | No clear effect |
| use FUMET without depth supervision versus supervised ZoeDepth on mixed data | FUMET · KITTI Eigen (832x512) | AbsRel: 0.120 -> 0.113 | Improved |
| add occlusion and truncation augmentations when training LSP (ablation) | LSP · KITTI car height | absolute relative error: 0.081 -> 0.062 | Improved |
| predict car width and length as well as height in LSP (ablation) | LSP · KITTI car height (All) | absolute relative error: 0.087 -> 0.062 | Improved |
Closest papers by meaning
- Towards Robust Driving Perception: A Flexible Scale-Driven Family for Self-Supervised Monocular Depth Estimation (ECCV 2026 · similarity 0.87)
- MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sources (ECCV 2026 · similarity 0.86)
- MGNet: Monocular Geometric Scene Understanding for Autonomous Driving (ICCV 2021 · similarity 0.86)
- Self-Supervised Relative Depth Learning for Urban Scene Understanding (ECCV 2018 · similarity 0.86)
- ViiNeuS: Volumetric Initialization for Implicit Neural Surface Reconstruction of Urban Scenes with Limited Image Overlap (CVPR 2025 · similarity 0.85)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (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.
