Research map / 3D vision / Depth and stereo
Supervising the new with the old: learning SFM from SFM
Maria Klodt, Andrea Vedaldi
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 SSIM structural-similarity image matching to the brightness constancy loss (ablation) | SfMLearner (reproduced) · KITTI Eigen split depth | abs. rel.: 0.198 -> 0.181 | Improved |
| add SSIM image matching (ablation) | SfMLearner (reproduced) · KITTI Eigen split depth | delta < 1.25: 0.732 -> 0.763 | Improved |
| add photometric uncertainty (probabilistic loss) (ablation) | SfMLearner + image matching · KITTI Eigen split depth | abs. rel.: 0.181 -> 0.180 | Improved |
| add photometric uncertainty (probabilistic loss) (ablation) | SfMLearner + image matching · KITTI Eigen split depth | RMSE: 6.771 -> 6.855 | Hurt |
| add photometric uncertainty (probabilistic loss) (ablation) | SfMLearner + image matching · KITTI Eigen split depth | sq. rel.: 2.054 -> 1.970 | Improved |
| add pose supervision from SFM (ORB-SLAM2) (ablation) | image matching + photometric uncertainty · KITTI Eigen split depth | abs. rel.: 0.180 -> 0.171 | Improved |
| add depth supervision from SFM on top of pose supervision (ablation) | + pose from SFM · KITTI Eigen split depth | sq. rel.: 1.891 -> 1.490 | Improved |
| add depth supervision from SFM on top of pose supervision (ablation) | + pose from SFM · KITTI Eigen split depth | RMSE: 6.588 -> 5.998 | Improved |
| pre-train on Cityscapes then fine-tune on KITTI instead of KITTI only | full method · KITTI Eigen split depth | abs. rel.: 0.166 -> 0.165 | Improved |
| train on Cityscapes only instead of KITTI | full method · KITTI Eigen split depth | abs. rel.: 0.166 -> 0.254 | Hurt |
| train on Virtual KITTI only instead of KITTI | full method · KITTI Eigen split depth | abs. rel.: 0.166 -> 0.270 | Hurt |
| full method vs SfMLearner (website) on odometry | proposed method · KITTI odometry Seq. 09 | ATE: 0.016 -> 0.014 | Improved |
Closest papers by meaning
- SfM-TTR: Using Structure From Motion for Test-Time Refinement of Single-View Depth Networks (CVPR 2023 · similarity 0.89)
- MP-SfM: Monocular Surface Priors for Robust Structure-from-Motion (CVPR 2025 · similarity 0.88)
- DeepSFM: Structure From Motion Via Deep Bundle Adjustment (ECCV 2020 · similarity 0.88)
- VGGSfM: Visual Geometry Grounded Deep Structure From Motion (CVPR 2024 · similarity 0.87)
- Learning Structure-from-Motion with Graph Attention Networks (CVPR 2024 · similarity 0.87)
- Light3R-SfM: Towards Feed-forward Structure-from-Motion (CVPR 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.
