Research map / 3D vision / Depth and stereo
Deep Depth from Focal Stack with Defocus Model for Camera-Setting Invariance
Yuki Fujimura, Masaaki Iiyama, Takuya Funatomi, Yasuhiro Mukaigawa
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 cost volume from defocus model instead of direct focal stack input | Ours vs DefocusNet · NYU Depth V2 blurred (train DefocusNet) | RMSE: 0.621 -> 0.186 | Improved |
| use cost volume under different train/test focus distances | Ours vs DefocusNet · DefocusNet dataset | RMSE: 0.299 -> 0.242 | Improved |
| add normalization to cost volume (ablation) | Ours (outlier removal only) · DefocusNet dataset sigma=1.0 | RMSE: 0.261 -> 0.239 | Improved |
| add outlier removal to normalized cost volume (ablation) | Ours (normalization only) · DefocusNet dataset sigma=9.0 | RMSE: 0.422 -> 0.403 | Improved |
| add outlier removal to normalized cost volume (ablation) | Ours (normalization only) · DefocusNet dataset sigma=1.0 | RMSE: 0.232 -> 0.239 | Hurt |
| add normalization to cost volume at large depth scale (ablation) | Ours (outlier removal only) · DefocusNet dataset sigma=9.0 | RMSE: 0.475 -> 0.403 | Improved |
| compensate scale by rescaling baseline outputs | DefocusNet · NYU Depth V2 blurred | RMSE: 0.621 -> 0.322 | Improved |
| train on NYU Depth V2 instead of DefocusNet dataset | Ours · NYU Depth V2 blurred | RMSE: 0.186 -> 0.054 | Improved |
| use explicit defocus model on DefocusNet dataset where defocus cues are short-range | Ours vs AiFDepthNet · DefocusNet dataset | RMSE: 0.156 -> 0.239 | Hurt |
| train and test on NYU Depth V2 with matched camera settings | Ours vs DefocusNet · NYU Depth V2 blurred (train NYU) | RMSE: 0.029 -> 0.054 | Hurt |
| add neighboring pixels to standard deviation cost | Ours · real captured data | depth noise | Improved |
Closest papers by meaning
- Bridging Unsupervised and Supervised Depth From Focus via All-in-Focus Supervision (ICCV 2021 · similarity 0.91)
- Learning Depth from Focus in the Wild (ECCV 2022 · similarity 0.91)
- Fully Self-Supervised Depth Estimation From Defocus Clue (CVPR 2023 · similarity 0.91)
- Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 2025 · similarity 0.89)
- Blurry-Edges: Photon-Limited Depth Estimation from Defocused Boundaries (CVPR 2025 · similarity 0.87)
- DualFocus: Depth from Focus with Spatio-Focal Dual Variational Constraints (NeurIPS 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.
