Research map / Label-efficient and robust learning / Self-supervised and contrastive
Fully Self-Supervised Depth Estimation From Defocus Clue
Read the paperCheck 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 |
|---|---|---|---|
| Use fully self-supervised focal-stack reconstruction without depth or AIF ground-truth | DAIF-Net · NYUv2 test set | δ1: Ours 0.950; DFF-DFV 0.967 | Mixed |
| Use fully self-supervised focal-stack reconstruction without depth or AIF ground-truth | DAIF-Net · DefocusNet test set, <0.5 m | δ1: Ours 0.889; DFF-FV 0.977 | Mixed |
| Add coarse AIF blurriness regularization (ablation) | DAIF-Net · NYUv2 test set | δ1: 0.934 -> 0.950 | Helped |
| Add depth smoothness regularization (ablation) | DAIF-Net · NYUv2 test set | δ1: 0.912 -> 0.950 | Helped |
| Add coarse AIF blurriness regularization (ablation) | DAIF-Net · NYUv2 test set | RMSE: 0.462 -> 0.325 | Helped |
| Add depth smoothness regularization (ablation) | DAIF-Net · NYUv2 test set | RMSE: 0.477 -> 0.325 | Helped |
| Train using a different set of camera parameters (ablation) | DAIF-Net · NYUv2 test set | δ1: 0.950 -> 0.957 | No clear effect |
| Add Gaussian noise to rendered focal-stack images (ablation) | DAIF-Net · Noisy NYUv2 test set | δ1: 0.950 -> 0.943 | No clear effect |
| Randomly sample five images from bins of a dense focal stack during training (ablation) | DAIF-Net · NYUv2 focal stacks with varying sizes and focus distances | depth estimation performance | Mixed |
| Increase the reconstruction-loss scale in loss-scale ablations (ablation) | DAIF-Net · NYUv2 test set | δ1: 0.904 -> 0.934 | Helped |
| Use the tested combination of reconstruction loss 100 and coarse AIF blur loss 10 (ablation) | DAIF-Net · NYUv2 test set | δ1: 0.950 | Helped |
Closest papers by meaning
- Single Image Depth Estimation Trained via Depth From Defocus Cues (CVPR 2019 · similarity 0.90)
- Focus on Defocus: Bridging the Synthetic to Real Domain Gap for Depth Estimation (CVPR 2020 · similarity 0.88)
- Blurry-Edges: Photon-Limited Depth Estimation from Defocused Boundaries (CVPR 2025 · similarity 0.88)
- Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 2025 · similarity 0.87)
- DualFocus: Depth from Focus with Spatio-Focal Dual Variational Constraints (NeurIPS 2025 · similarity 0.86)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 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.