AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Gaussian Kernel Mixture Network for Single Image Defocus Deblurring

Yuhui Quan, Zicong Wu, Hui Ji

NeurIPS 2021 · Architectures and efficiency › CNN design · approach: Dense matching and estimation CNNs · 23 citations (OpenAlex)

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 changedWhereMeasuredEffect
use multi-scale (T=3) estimation instead of single scale (ablation)GKMNet · DPD test setPSNR (dB): 23.63 -> 25.46Improved
add Conv-LSTM in the attention prediction unit (ablation)GKMNet · DPD test setPSNR (dB): 25.00 -> 25.46Improved
wrap SRAM in GKM unrolled fixed-point structure instead of using it as a plain end-to-end network (ablation)GKMNet · DPD test setPSNR (dB): 21.98 -> 25.46Improved
learn GCM kernels initialized from Gaussian kernels (ablation)GKMNet · DPD test setPSNR (dB): 25.46 -> 25.49No clear effect
learn GCM kernels with random initialization (ablation)GKMNet · DPD test setPSNR (dB): 25.46 -> 25.35Hurt
use 3 iterations at a single scale (S=1,R=3) instead of S=3,R=1 (ablation)GKMNet · DPD test setPSNR (dB): 25.47 -> 25.11Hurt
add a second iteration per scale (S=3,R=2) (ablation)GKMNet · DPD test setPSNR (dB): 25.47 -> 25.53 PSNR; SSIM 0.789 -> 0.773Mixed
use GKMNet single-image model versus dual-pixel two-view DPDNet-DGKMNet · DPD test setPSNR (dB): 25.12 -> 25.47Improved
use GKMNet instead of AttNet under heavy noiseRTF noise 2.55 · RTFPSNR (dB): 23.83 -> 25.01Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy