Research map / 3D vision / Depth and stereo
Bridging Unsupervised and Supervised Depth From Focus via All-in-Focus Supervision
Ning-Hsu Wang, Ren Wang, Yu-Lun Liu, Yu-Hao Huang, Yu-Lin Chang, Chia-Ping Chen, Kevin Jou
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 3D convolution instead of 2D convolution (ablation) | Depth-from-focus attention network · 4D Light Field Dataset | MAE: 0.1070 -> 0.0788 | Improved |
| use 3D convolution instead of 2D convolution (ablation) | Depth-from-focus attention network · 4D Light Field Dataset | Bumpiness: 2.1435 -> 1.5776 | Improved |
| add the shared attention mechanism (ablation) | Depth-from-focus attention network · 4D Light Field Dataset (3D) | MAE: 0.0851 -> 0.0788 | Improved |
| add the shared attention mechanism (ablation) | Depth-from-focus attention network · 4D Light Field Dataset (3D) | MSE: 0.0461 -> 0.0472 | No clear effect |
| add the shared attention mechanism (ablation) | Depth-from-focus attention network · 4D Light Field Dataset (3D) | Bumpiness: 2.3168 -> 1.5776 | Improved |
| add smoothness loss to unsupervised training (stack 10) (ablation) | Depth-from-focus attention network · 4D Light Field Dataset | MAE: 0.2425 -> 0.1671 | Improved |
| add smoothness loss to unsupervised training (stack 37) (ablation) | Depth-from-focus attention network · 4D Light Field Dataset | MAE: 0.2099 -> 0.1116 | Improved |
| increase focal stack size from 10 to 37 in unsupervised training (ablation) | Depth-from-focus attention network · 4D Light Field Dataset | MAE: 0.1671 -> 0.1116 | Improved |
| use supervised depth instead of AiF supervision (stack 10, no smoothness) (ablation) | Depth-from-focus attention network · 4D Light Field Dataset | MAE: 0.2425 -> 0.0788 | Improved |
| train with arbitrary stack sizes instead of fixed size 10 (ablation) | Depth-from-focus attention network · 4D Light Field Dataset, varying test stack size | disparity error | Improved |
| use supervised ours instead of DefocusNet | Depth-from-focus attention network · 4D Light Field Dataset | MSE: 0.0593 -> 0.0472 | Improved |
| use supervised ours instead of DefocusNet | Depth-from-focus attention network · DefocusNet dataset | MAE: 0.0637 -> 0.0549 | Improved |
Closest papers by meaning
- Fully Self-Supervised Depth Estimation From Defocus Clue (CVPR 2023 · similarity 0.92)
- Deep Depth from Focal Stack with Defocus Model for Camera-Setting Invariance (IJCV 2023 · similarity 0.91)
- Single Image Depth Estimation Trained via Depth From Defocus Cues (CVPR 2019 · similarity 0.91)
- Learning Depth from Focus in the Wild (ECCV 2022 · similarity 0.91)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.88)
- Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 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.
