Research map / 3D vision / Depth and stereo
CAM-Convs: Camera-Aware Multi-Scale Convolutions for Single-View Depth
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 camera-aware convolutions using camera intrinsics | U-Net encoder-decoder · 2D-3D Semantics, s1 U FocalR-f128 | abs.rel: 0.189 -> 0.175 | Helped |
| Add camera-aware convolutions using camera intrinsics | U-Net encoder-decoder · 2D-3D Semantics, s2 U FocalR-f128 | abs.rel: 0.166 -> 0.158 | Helped |
| Add camera-aware convolutions using camera intrinsics | U-Net encoder-decoder · 2D-3D Semantics, s3 U FocalR-f128 (unseen sensor size) | abs.rel: 0.174 -> 0.164 | Helped |
| Add camera-aware convolutions using camera intrinsics | U-Net encoder-decoder · 2D-3D Semantics, s5 f64 (unseen camera) | rmse: 0.309 -> 0.289 | Helped |
| Train at a different focal length instead of matching the test camera | Standard U-Net · 2D-3D Semantics, s1 f64 | l1.inv: 0.184 -> 0.318 | Hurt |
| Use multi-focal training with focal length normalization | Standard U-Net · 2D-3D Semantics, s1 f64 | l1.inv: 0.659 -> 0.189 | Helped |
| Remove focal length normalization from multi-focal training (ablation) | Standard U-Net · 2D-3D Semantics, s1 f64 | l1.inv: 0.189 -> 0.659 | Hurt |
| Train with multiple sensor sizes using weight sharing | Standard U-Net · 2D-3D Semantics, s1 U FocalR-f128 | rmse: 0.46 -> 0.484 | Hurt |
| Resize images to a common sensor size during training | FCN · KITTI, sK | abs.rel: 9.16 -> 9.08 | Helped |
| Resize images to a common sensor size during training | FCN · ScanNet, sS | abs.rel: 0.12 -> 0.12 | No clear effect |
| Train on multiple datasets without camera-aware convolutions (ablation) | Multi-dataset depth network · KITTI, ScanNet, MegaDepth and Sun3D training | convergence | Hurt |
Closest papers by meaning
- Single Image Depth Estimation Trained via Depth From Defocus Cues (CVPR 2019 · similarity 0.89)
- Robust Consistent Video Depth Estimation (CVPR 2021 · similarity 0.89)
- Learning Depth From Monocular Videos Using Direct Methods (CVPR 2018 · similarity 0.89)
- Camera Pose Matters: Improving Depth Prediction by Mitigating Pose Distribution Bias (CVPR 2021 · similarity 0.88)
- Video Depth Anything: Consistent Depth Estimation for Super-Long Videos (CVPR 2025 · similarity 0.86)
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (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.