Research map / Image generation / Image restoration
Exploring Positional Characteristics of Dual-Pixel Data for Camera Autofocus
Myungsub Choi, Hana Lee, Hyong-euk Lee
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 |
|---|---|---|---|
| add RoI position encoding (RoI-PE) to the dual-pixel AF model | MobileNet-v2 AF model, single slice · Dual-pixel AF dataset (D1) | accuracy exact (=0): 0.164 -> 0.240 | Improved |
| add RoI-PE to full-stack model | MobileNet-v2 AF model, full stack · Dual-pixel AF dataset (D*) | accuracy within 4 focal indices: 0.955 -> 0.972 | Improved |
| add lens-PE channel instead of 98 zero-masked channels (ablation) | MobileNet-v2 AF model · Dual-pixel AF dataset (D1) | accuracy exact (=0): 0.068 -> 0.182 | Improved |
| replace 98-channel input and 4.0 MobileNet with lens-PE and 1.0 MobileNet | AF model · Computational complexity | # of params: 34.75M -> 2.30M | Improved |
| replace MobileNet with MCUNet backbone (ablation) | AF model with PEs · Dual-pixel AF dataset (D1) | accuracy exact (=0): 0.240 -> 0.210 | Mixed |
| efficiency gains on runtime | AF model with PEs · Computational complexity | Runtime: 6.42ms -> 6.22ms | No clear effect |
| use polar coordinates instead of (x, y) for RoI-PE (ablation) | RoI-PE · Dual-pixel AF dataset (D1) | MAE: 1.803 -> 1.850 | Hurt |
| use radius-only RoI-PE (ablation) | RoI-PE · Dual-pixel AF dataset (D1) | MAE: 1.803 -> 2.079 | Hurt |
| add RoI-PE to conventional single-channel raw images | AF model · Conventional raw image (I1) | accuracy exact (=0): 0.115 -> 0.131 | Improved |
| add second step with temperature T=0.5 ordinal loss | multi-step AF model · Dual-pixel AF dataset (D1) | MAE: 1.803 -> 1.668 | Improved |
| lower temperature from 0.5 to 0.33 in ordinal loss (ablation) | multi-step AF model · Dual-pixel AF dataset (D1) | MAE: 1.668 -> 1.674 | No clear effect |
Closest papers by meaning
- Learning to Autofocus (CVPR 2020 · similarity 0.86)
- DualFocus: Depth from Focus with Spatio-Focal Dual Variational Constraints (NeurIPS 2025 · similarity 0.85)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.84)
- Stabilizing and Accelerating Autofocus with Expert Trajectory Regularized Deep Reinforcement Learning (CVPR 2025 · similarity 0.84)
- A Physics-Informed Blur Learning Framework for Imaging Systems (CVPR 2025 · similarity 0.82)
- One-Step Event-Driven High-Speed Autofocus (CVPR 2025 · similarity 0.82)
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.
