Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Domain Adaptation for Depth Prediction from Images
Alessio Tonioni, Matteo Poggi, Stefano Mattoccia, Luigi Di Stefano
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 |
|---|---|---|---|
| fine-tune on raw AD disparities without confidence (ablation) | DispNetC · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 10.86 -> 11.73 | Hurt |
| add confidence weighting (tau=0) (ablation) | Regression · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 11.73 -> 3.66 | Improved |
| add masking at tau=0.8 over weighting (ablation) | Weighted · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 3.66 -> 3.17 | Improved |
| add smoothness loss (ablation) | Masked · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 3.17 -> 3.17 | No clear effect |
| add smoothness loss (ablation) | Masked · KITTI raw stereo, similar domains | bad3 (%): 3.97 -> 3.78 | Improved |
| add reprojection loss (ablation) | Masked · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 3.17 -> 3.03 | Improved |
| add both smoothness and reprojection (complete) (ablation) | Masked · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 3.17 -> 2.96 | Improved |
| learn tau instead of fixed 0.8 (ablation) | Adaptation-AD · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 2.96 -> 3.15 | Hurt |
| use SGM instead of AD for labels (ablation) | Adaptation · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 2.96 -> 2.58 | Improved |
| fuse AD and SGM labels (ablation) | Adaptation-SGM · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 2.58 -> 2.61 | No clear effect |
| unsupervised adaptation versus GT tuning on few frames | Adaptation-SGM · KITTI raw stereo, target domain (4-fold CV) | bad3 (%): 5.04 -> 2.58 | Improved |
| add confidence masking over raw regression for mono (ablation) | monodepth ResNet50+pp AD · KITTI Eigen split monodepth | Abs Rel: 0.230 -> 0.109 | Improved |
Closest papers by meaning
- 3D-PL: Domain Adaptive Depth Estimation with 3D-Aware Pseudo-Labeling (ECCV 2022 · similarity 0.91)
- AdaDepth: Unsupervised Content Congruent Adaptation for Depth Estimation (CVPR 2018 · similarity 0.91)
- Unsupervised Adaptation for Deep Stereo (ICCV 2017 · similarity 0.91)
- Geometry-Aware Symmetric Domain Adaptation for Monocular Depth Estimation (CVPR 2019 · similarity 0.91)
- Domain Decluttering: Simplifying Images to Mitigate Synthetic-Real Domain Shift and Improve Depth Estimation (CVPR 2020 · similarity 0.90)
- DESC: Domain Adaptation for Depth Estimation via Semantic Consistency (IJCV 2022 · similarity 0.90)
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.
