Research map / 3D vision / Depth and stereo
SfM-TTR: Using Structure From Motion for Test-Time Refinement of Single-View Depth Networks
Sergio Izquierdo, Javier Civera
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 SfM sparse depths for test-time refinement | AdaBins · KITTI Eigen test split, new ground truth, no cropping | Abs Rel: 0.072 -> 0.060 | Improved |
| Use SfM sparse depths for test-time refinement | ManyDepth · KITTI Eigen test split, new ground truth, no cropping | RMSE: 3.116 -> 2.648 | Improved |
| Use SfM sparse depths for test-time refinement | CADepth · KITTI Eigen test split, new ground truth, no cropping | Abs Rel: 0.078 -> 0.068 | Improved |
| Use SfM sparse depths for test-time refinement | DIFFNet · KITTI Eigen test split, new ground truth, no cropping | RMSE: 3.230 -> 2.600 | Improved |
| Replace photometric refinement with SfM-TTR | CADepth · KITTI Eigen test split, new ground truth, no cropping | Abs Rel: 0.088 -> 0.068 | Improved |
| Replace photometric refinement with SfM-TTR | ManyDepth · KITTI Eigen test split, new ground truth, no cropping | RMSE: 3.034 -> 2.648 | Improved |
| Use SfM-TTR instead of no refinement | ManyDepth · KITTI Eigen test split, new ground truth, no cropping | Abs Rel: 0.064 -> 0.057; Ph-TTR: 0.064 -> 0.056 | Mixed |
| Use median scale alignment (ablation) | AdaBins + SfM-TTR · KITTI alignment ablation | Abs Rel: 0.060 -> 0.074 | Hurt |
| Use strict RANSAC inlier sets rather than relaxed supervision inliers (ablation) | AdaBins + SfM-TTR · KITTI alignment ablation | Abs Rel: 0.060 -> 0.065 | Hurt |
| Use unweighted least-squares scale alignment (ablation) | AdaBins + SfM-TTR · KITTI alignment ablation | Abs Rel: 0.060 -> 0.064 | Hurt |
| Set reconstruction reliability weights to 1 (ablation) | AdaBins + SfM-TTR · KITTI alignment ablation | Abs Rel: 0.060 -> 0.062 | Hurt |
| Fine-tune only the encoder rather than the full network (ablation) | AdaBins + SfM-TTR · KITTI encoder versus full-network refinement | Abs Rel: 0.062 -> 0.060 | Improved |
Closest papers by meaning
- Multi-view Reconstruction via SfM-guided Monocular Depth Estimation (CVPR 2025 · similarity 0.88)
- MP-SfM: Monocular Surface Priors for Robust Structure-from-Motion (CVPR 2025 · similarity 0.87)
- VGGSfM: Visual Geometry Grounded Deep Structure From Motion (CVPR 2024 · similarity 0.87)
- Light3R-SfM: Towards Feed-forward Structure-from-Motion (CVPR 2025 · similarity 0.87)
- ColabSfM: Collaborative Structure-from-Motion by Point Cloud Registration (CVPR 2025 · similarity 0.87)
- Detector-Free Structure from Motion (CVPR 2024 · 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.
