Research map / Label-efficient and robust learning / Domain adaptation and generalization
3D-PL: Domain Adaptive Depth Estimation with 3D-Aware Pseudo-Labeling
Yu-Ting Yen, Chia-Ni Lu, Wei-Chen Chiu, Yi-Hsuan Tsai
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 consistency-masked 2D pseudo-labels (ablation) | depth model (synthetic only) · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.771 -> 1.095 | Improved |
| add 3D-aware completion pseudo-labels (ablation) | depth model (synthetic only) · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.771 -> 1.054 | Improved |
| add consistency mask to 2D pseudo-labels (ablation) | y_cons pseudo-labels · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.125 -> 1.095 | Improved |
| restrict 3D completion input to confident pixels (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.070 -> 1.049 | Improved |
| combine 2D and 3D-aware pseudo-labels (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.095 -> 1.049 | Improved |
| use completion label as main reference over consistency label (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.095 -> 1.049 | Improved |
| replace 3D completion with 2D depth completion trained on vKITTI (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.054 -> 1.476 | Hurt |
| add synthetic task loss alongside pseudo-label loss (alpha 0.7 vs 1) (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Abs Rel: 0.191 -> 0.162 | Improved |
| vary pseudo-label threshold tau from 0.1 to 3 (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.048 -> 1.066 | No clear effect |
| use L_task^s only instead of both L_task^s and L_task^{s->r} (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.074 -> 1.049 | Improved |
| increase loss weights lambda_cons, lambda_comp from (1,0.1) to (10,10) (ablation) | 3D-PL · KITTI Eigen test (50m cap, single image) ablation | Sq Rel: 1.049 -> 1.138 | Hurt |
| add 3D-PL to stereo training vs SharinGAN | stereo baseline · KITTI Eigen test (50m cap, stereo) | Sq Rel: 0.673 -> 0.641 | Improved |
Closest papers by meaning
- Unsupervised Domain Adaptation for Depth Prediction from Images (TPAMI 2019 · similarity 0.91)
- DESC: Domain Adaptation for Depth Estimation via Semantic Consistency (IJCV 2022 · similarity 0.91)
- Geometry-Aware Symmetric Domain Adaptation for Monocular Depth Estimation (CVPR 2019 · similarity 0.90)
- Pseudo Label Refinery for Unsupervised Domain Adaptation on Cross-dataset 3D Object Detection (CVPR 2024 · similarity 0.88)
- Depth Anything V2 (NeurIPS 2024 · similarity 0.87)
- A Simple yet Universal Framework for Depth Completion (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.
