Research map / 3D vision / Depth and stereo
Camera Pose Matters: Improving Depth Prediction by Mitigating Pose Distribution Bias
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 perspective-aware data augmentation (PDA) and camera pose prior encoding (CPP) | UNet depth predictor · InteriorNet Uniform-Test-Set | Absᵣ: .183 -> .091 | Helped |
| Add PDA and CPP | UNet depth predictor · ScanNet Uniform-Test-Set | Absᵣ: .177 -> .143 | Helped |
| Add PDA to geometrically reproject training views (ablation) | UNet depth predictor · InteriorNet Uniform-Test-Set | Absᵣ: .183 -> .168 | Helped |
| Add CPP encoding of camera pose (ablation) | UNet depth predictor · InteriorNet Uniform-Test-Set | Absᵣ: .183 -> .106 | Helped |
| Resample uncommon-pose training examples instead of adding new views (ablation) | UNet depth predictor · InteriorNet Uniform-Test-Set | Absᵣ: .183 -> .210 | Hurt |
| Use random-crop conventional augmentation instead of no added augmentation (ablation) | UNet depth predictor · ScanNet Uniform-Test-Set | Absᵣ: .177 -> .174 | Helped |
| Use geometric PDA rather than conventional random-crop augmentation (ablation) | UNet depth predictor · InteriorNet Uniform-Test-Set | Absᵣ: .172 -> .168 | Helped |
| Encode pose as constant-value channels rather than CPP maps (ablation) | UNet depth predictor · InteriorNet Uniform-Test-Set | Absᵣ: .137 -> .106 | Hurt |
| Train on narrow pose ranges and add PDA plus CPP | UNet depth predictor · InteriorNet Natural-Test-Set (OOD) | Absᵣ: .265 -> .216 | Helped |
| Use CPP with predicted rather than true camera poses (ablation) | PDA+CPP depth predictor · ScanNet Uniform-Test-Set, synthetic-to-real | Absᵣ: .245 -> .273 | Hurt |
| Use conventional random-crop augmentation (ablation) | synthetically trained depth predictor · ScanNet Uniform-Test-Set, synthetic-to-real | Absᵣ: .305 -> .316 | Hurt |
| Replace inverse-tangent CPP encoding with clipping (ablation) | CPP depth predictor · ScanNet Uniform-Test-Set | Absᵣ: .154 -> .157 | Hurt |
Closest papers by meaning
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.89)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.88)
- Relative Pose Estimation through Affine Corrections of Monocular Depth Priors (CVPR 2025 · similarity 0.87)
- DME: Unveiling the Bias for Better Generalized Monocular Depth Estimation (AAAI 2024 · similarity 0.87)
- One View Is Enough: In-the-Wild Monocular Pretraining for Novel View Generation (NeurIPS 2026 · similarity 0.87)
- Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priors (CVPR 2025 · 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.