AI Paper Atlas

Research map / Image generation / Super-resolution

SRWarp: Generalized Image Super-Resolution under Arbitrary Transformation

Sanghyun Son, Kyoung Mu Lee

CVPR 2021 · Image generation › Super-resolution · approach: CNN and transformer SR architectures · 38 citations (OpenAlex)

Check 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 SRWarp instead of ×4 SR followed by bicubic warpingimage warping · DIV2KW TestmPSNR (dB): 30.76 -> 31.04Improved
Use SRWarp for arbitrary-scale SRB100 · scale ×2.0 to ×4.0PSNR (dB): 32.33 vs. 32.31; 31.45 -> 31.46; 30.46 -> 30.46; 29.69 -> 29.71; 29.26 -> 29.27; 28.88 -> 28.89; 28.41 -> 28.42; 28.01 -Mixed
Add adaptive warping, multiscale blending, and reconstruction (ablation)MDSR baseline · DIV2KW Valid.mPSNR (dB): 31.36 -> 32.29Improved
Add multiscale blending (ablation)MDSR with AWL and reconstruction · DIV2KW Valid.mPSNR (dB): 32.19 -> 32.29Improved
Add reconstruction module (ablation)MDSR with AWL and multiscale blending · DIV2KW Valid.mPSNR (dB): 32.21 -> 32.29Improved
Add adaptive warping layer (ablation)MDSR with multiscale blending and reconstruction · DIV2KW Valid.mPSNR (dB): 32.19 -> 32.29Improved
Use adaptive coordinates with fixed cubic-spline weights (ablation)SRWarp warping layer · DIV2KW Valid.mPSNR (dB): 32.19No clear effect
Use learned kernels on a regular grid (ablation)SRWarp warping layer · DIV2KW Valid.mPSNR (dB): 32.19 -> 32.25Improved
Combine adaptive coordinates with a learned kernel estimator (ablation)SRWarp warping layer · DIV2KW Valid.mPSNR (dB): 32.25 -> 32.29Improved
Use scale-specific kernel estimators (ablation)AWL · DIV2KW Valid.mPSNR (dB): +0.01dBImproved
Estimate resampling kernels depthwise (ablation)AWL · DIV2KW Valid.mPSNR (dB): +0.05dBImproved
Average warped features across scales instead of using the proposed blending (ablation)SRWarp multiscale blending · DIV2KW Valid.mPSNR (dB): 32.26 -> 32.29Hurt

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