Research map / Image generation / Image restoration
Learning To Reduce Defocus Blur by Realistically Modeling Dual-Pixel Data
Abdullah Abuolaim, Mauricio Delbracio, Damien Kelly, Michael S. Brown, Peyman Milanfar
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 the proposed parametric Butterworth DP-PSF model instead of the prior PSF model (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 24.69 -> 25.39 | Improved |
| add radial distortion to synthetic DP data (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 25.19 -> 25.39 | Improved |
| add the right DP view (dual views instead of a single view) (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 24.24 -> 25.39 | Improved |
| add radial distance patch as input (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 25.00 -> 25.39 | Improved |
| add three-scale edge loss versus no edge loss (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 25.06 -> 25.39 | Improved |
| add three-scale edge loss instead of single-scale 3x3 Sobel loss (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 25.11 -> 25.39 | Improved |
| add single-scale Sobel loss versus no edge loss (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 25.06 -> 25.11 | No clear effect |
| add both radial distance patch and multi-scale edge loss (ablation) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 24.90 -> 25.39 | Improved |
| train with image sequences (multi-frame) instead of single frames (ablation) | RDPD+ · Synthetic DP data test | PSNR: 30.26 -> 31.09 | Improved |
| add synthetic DP data to training of DPDNet | DPDNet · Synthetic DP data test | PSNR: 26.38 -> 29.84 | Improved |
| add synthetic DP data to training of DPDNet on outdoor real scenes | DPDNet · Canon DP test (outdoor) | PSNR: 22.90 -> 22.72 | Hurt |
| add synthetic DP data to training of RDPD (RDPD+ vs DPDNet) | RDPD+ · Canon DP test (indoor and outdoor) | PSNR: 25.13 -> 25.39 | Improved |
Closest papers by meaning
- Defocus Deblurring Using Dual-Pixel Data (ECCV 2020 · similarity 0.93)
- Learning to Deblur Using Light Field Generated and Real Defocus Images (CVPR 2022 · similarity 0.92)
- Quad-Pixel Image Defocus Deblurring: A New Benchmark and Model (CVPR 2025 · similarity 0.91)
- LDP: Language-driven Dual-Pixel Image Defocus Deblurring Network (CVPR 2024 · similarity 0.89)
- Residual Diffusion Deblurring Model for Single Image Defocus Deblurring (AAAI 2025 · similarity 0.89)
- Prior and Prediction Inverse Kernel Transformer for Single Image Defocus Deblurring (AAAI 2024 · similarity 0.88)
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.
