AI Paper Atlas

Research map / Image generation / Editing and style transfer

Exploiting Semantics for Face Image Deblurring

Ziyi Shen, Wei‐Sheng Lai, Tingfa Xu, Jan Kautz, Ming–Hsuan Yang

IJCV 2020 · Image generation › Editing and style transfer · approach: Identity-preserving face manipulation · 43 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
add ground-truth semantic labels as input (ablation)coarse deblurring network G · HelenPSNR: 24.85 -> 25.85Improved
extract labels from coarse deblurred image instead of blurred image (ablation)fine deblurring with parsing · HelenPSNR: 25.32 -> 25.48Improved
extract labels from coarse deblurred image instead of blurred image (ablation)face parsing network · Helenparsing F-score: 0.615 -> 0.628Improved
add equal-weight local structural loss (ablation)content-loss fine network · HelenPSNR: 25.48 -> 25.72Improved
use adaptive instead of equal weights in local structural loss (ablation)content-loss fine network · HelenPSNR: 25.72 -> 25.80Improved
use adaptive instead of equal weights in local structural loss (ablation)content-loss fine network · CelebASSIM: 0.869 -> 0.874Improved
train stage-by-stage plus joint fine-tuning instead of from scratch (ablation)full three-network model · HelenPSNR: 24.74 -> 25.92Improved
add perceptual (VGG) loss (ablation)full model · HelenPSNR: 25.92 -> 26.28Improved
add adversarial loss on top of perceptual loss (ablation)full model · HelenSSIM: 0.877 -> 0.876No clear effect
enlarge first conv kernel from 5x5 to 11x11 (ablation)deblurring networks · HelenPSNR: 25.63 -> 25.80Improved
enlarge first conv kernel from 11x11 to 13x13 (ablation)deblurring networks · HelenPSNR: 25.80 -> 25.80No clear effect
add a second fine deblurring stage (ablation)full model · HelenPSNR: 25.80 -> 25.87No clear effect

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