Research map / 3D vision / Depth and stereo
Learning Single Camera Depth Estimation Using Dual-Pixels
Rahul Garg, Neal Wadhwa, Sameer Ansari, Jonathan T. Barron
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 affine-invariant loss instead of scale-invariant loss (ablation) | DPNet (RGB+DP input) · Dual-pixel dataset (our depth) | geometric mean: .0477 -> .0422 | Improved |
| use affine-invariant loss instead of no invariance (ablation) | DPNet (RGB+DP input) · Dual-pixel dataset (our depth) | geometric mean: .1530 -> .0422 | Improved |
| use affine-invariant loss instead of scale-invariant loss (ablation) | VGG (RGB+DP input) · Dual-pixel dataset (our depth) | geometric mean: .0535 -> .0446 | Improved |
| use affine-invariant loss instead of scale-invariant loss (ablation) | Godard et al. (ResNet50) · Dual-pixel dataset (our depth) | geometric mean: .0613 -> .0592 | Improved |
| use affine-invariant loss instead of scale-invariant loss (ablation) | Garg et al. (ResNet50) · Dual-pixel dataset (our depth) | geometric mean: .0613 -> .0584 | Improved |
| use affine-invariant loss instead of scale-invariant loss (ablation) | DPNet (RGB input) · Dual-pixel dataset (our depth) | geometric mean: .1047 -> .1014 | Improved |
| add dual-pixel data to RGB input (ablation) | DPNet with affine invariance · Dual-pixel dataset (our depth) | geometric mean: .1014 -> .0422 | Improved |
| use 3D assisted loss instead of folded loss (ablation) | DPNet · Dual-pixel dataset (our depth) | AIWE(1): .0225 -> .0175 | Improved |
| use 3D assisted loss instead of folded loss (ablation) | DPNet · Dual-pixel dataset (our depth) | 1 - |rho_s|: .195 -> .139 | Improved |
| train on all 4 alternative views instead of stereo views only (ablation) | DPNet · Dual-pixel dataset (our depth) | AIWE(1): .0218 -> .0175 | Improved |
| evaluate on photos from phones unseen in training (extended test set) vs standard test set | DPNet · Dual-pixel dataset (our depth) | AIWE(1): .0175 -> .0188 | Hurt |
| use small DPNet instead of larger VGG under affine invariance | DPNet vs VGG (RGB+DP) · Dual-pixel dataset (our depth) | geometric mean: .0446 -> .0422 | Improved |
Closest papers by meaning
- Du²Net: Learning Depth Estimation from Dual-Cameras and Dual-Pixels (ECCV 2020 · similarity 0.93)
- Spatio-Focal Bidirectional Disparity Estimation From a Dual-Pixel Image (CVPR 2023 · similarity 0.91)
- Single Image Depth Estimation Trained via Depth From Defocus Cues (CVPR 2019 · similarity 0.91)
- A Survey on Deep Learning Techniques for Stereo-Based Depth Estimation (TPAMI 2020 · similarity 0.90)
- Single View Stereo Matching (CVPR 2018 · similarity 0.90)
- Monocular Depth Estimation: A Thorough Review (TPAMI 2023 · 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.
