Research map / Image generation / Image restoration
K3DN: Disparity-Aware Kernel Estimation for Dual-Pixel Defocus Deblurring
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 disparity-aware kernel estimation with the full K3DN framework | K3DN · DPD-blur test | PSNR (dB): 26.66 -> 27.06 | Helped |
| Use disparity-aware kernel estimation with the full K3DN framework | K3DN · DDD-syn test | PSNR (dB): 36.53 -> 38.23 | Helped |
| Use disparity-aware kernel estimation with the full K3DN framework | K3DN · RDPD test | PSNR (dB): 32.27 -> 32.84 | Helped |
| Add reblur regularization (ablation) | K3DN · DPD-blur | PSNR (dB): 26.80 -> 26.84 | Helped |
| Add the SRP block to preserve in-focus features (ablation) | K3DN · DPD-blur | PSNR (dB): 26.76 -> 26.84 | Helped |
| Add the PSF block and reblur regularization (ablation) | K3DN · DPD-blur | PSNR (dB): 26.69 -> 26.84 | Helped |
| Add the SRP block and reblur regularization (ablation) | K3DN · DPD-blur | PSNR (dB): 26.68 -> 26.84 | Helped |
| Increase trainable kernel set size from 64 to 192 (ablation) | K3DN · DPD-blur | PSNR (dB): 26.71 -> 26.87 | Helped |
| Increase trainable kernel set size from 160 to 192 (ablation) | K3DN · DPD-blur | SSIM: 0.829 -> 0.830 | Helped |
Closest papers by meaning
- LDP: Language-driven Dual-Pixel Image Defocus Deblurring Network (CVPR 2024 · similarity 0.91)
- Prior and Prediction Inverse Kernel Transformer for Single Image Defocus Deblurring (AAAI 2024 · similarity 0.89)
- Quad-Pixel Image Defocus Deblurring: A New Benchmark and Model (CVPR 2025 · similarity 0.89)
- Parameterized Blur Kernel Prior Learning for Local Motion Deblurring (CVPR 2025 · similarity 0.88)
- Residual Diffusion Deblurring Model for Single Image Defocus Deblurring (AAAI 2025 · similarity 0.87)
- Real-World Efficient Blind Motion Deblurring via Blur Pixel Discretization (CVPR 2024 · 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.