AI Paper Atlas

Research map / Image generation / Super-resolution

Learning Re-sampling Methods with Parameter Attribution for Image Super-resolution

Xiaotong Luo, Yuan Xie, Yanyun Qu

NeurIPS 2023 · Image generation › Super-resolution · approach: CNN and transformer SR architectures · 1 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 inverse sampling to uniform sampling baseline (ablation)EDSR_baseline · Set14 4x SRPSNR: 28.22 -> 28.25Improved
alternate uniform and inverse sampling training (w/o IG) (ablation)EDSR_baseline · Set14 4x SRPSNR: 28.22 -> 28.28Improved
use filter-level integrated gradient parameter attribution (ablation)EDSR_baseline BSPA · Set14 4x SRPSNR: 28.28 -> 28.32Improved
use weight-level instead of filter-level IG (ablation)EDSR_baseline BSPA · Urban100 4x SRPSNR: 25.47 -> 25.53Improved
selective ratio beta from 1.0 down to 0.1 (ablation)EDSR_baseline BSPA · Urban100 4x SRPSNR: 25.43 -> 25.53Improved
unbalanced factor from 10 up to 200 (ablation)EDSR_baseline BSPA · Urban100 4x SRPSNR: 25.53 -> 25.30Hurt
number of classes 5 vs 10 vs 20 (ablation)EDSR_baseline BSPA · Urban100 4x SRPSNR: 25.53 -> 25.54No clear effect
apply BSPA to EDSR_baselineEDSR_baseline · Urban100 4x SRPSNR: 25.27 -> 25.53Improved
apply BSPA to RCANRCAN · Urban100 4x SRPSNR: 25.54 -> 26.02Improved
apply BSPA to FSRCNN (small model)FSRCNN · Set5 4x SRPSNR: 29.02 -> 29.04Hurt
compare BSPA with UDN loss reweighting on EDSR_baselineEDSR_baseline · Set5 4x SRPSNR: 31.44 -> 31.80Improved
BSPA vs SamplingAug on RCANRCAN · Manga109 4x SRPSNR: 30.17 -> 30.34Improved

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