AI Paper Atlas

Research map / Image generation / Super-resolution

Learning a Single Network for Scale-Arbitrary Super-Resolution

Longguang Wang, Yingqian Wang, Zaiping Lin, Jungang Yang, Wei An, Yulan Guo

ICCV 2021 · Image generation › Super-resolution · approach: CNN and transformer SR architectures · 171 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 scale-aware upsampling instead of bicubic upsampling (ablation)EDSR · Set5 x1.7/x2PSNR: 39.56/37.86 -> 39.76/38.13Improved
add scale-aware convolution to feature adaption (ablation)variant 2 · Set5 x1.3/x3.9PSNR: 34.32 -> 34.38Improved
add guidance map generation (ablation)variant 3 · Set5 x1.3/x3.9PSNR: 34.38 -> 34.44Improved
add scale-aware feature adaption (conv + guidance map) (ablation)variant 2 · Set5 x1.7/x2/x1.9/x3.5PSNR: 39.76/38.13/34.91 -> 39.87/38.19/35.03Improved
use 4 experts instead of 1 in scale-aware convolution (ablation)ArbEDSR · Set5 x1.7/x2/x2.55PSNR: 39.37/37.87/35.82 -> 39.88/38.22/36.03Improved
use 4 experts instead of 1 on highly asymmetric scale (ablation)ArbEDSR · Set5 x1.3/x3.9PSNR: 33.34 -> 34.44Improved
increase experts from 4 to 8 (ablation)ArbEDSR · Set5 x1.3/x3.9PSNR: 34.44 -> 34.46No clear effect
plug scale-aware module into EDSR vs bicubic-resized EDSRArbEDSR · Set5 x1.6/x1.55PSNR: 40.39/40.71 -> 40.64/40.94Improved
plug scale-aware module into EDSR at integer scale x2ArbEDSR · Set5 x2PSNR: 38.19 -> 38.19No clear effect
replace Meta-RCAN with ArbRCAN (scale-aware conv) on asymmetric SRMeta-RCAN+ft · Manga109 asymmetricPSNR: 37.80/33.67/34.28 -> 37.93/33.81/34.41Improved
fine-tune Meta-RCAN on the authors' training setMeta-RCAN · Manga109 asymmetricPSNR: 37.74/33.61/34.23 -> 37.80/33.67/34.28No clear effect
use ArbRCAN instead of Meta-RCAN for memory and timeMeta-RCAN · B100/LR 100x100 x4memory, time: 3.1G -> 1.1G; 0.39s -> 0.29sImproved

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