AI Paper Atlas

Research map / Image generation / Super-resolution

Finding Discriminative Filters for Specific Degradations in Blind Super-Resolution

Liangbin Xie, Xintao Wang, Chao Dong, Zhongang Qi, Ying Shan

NeurIPS 2021 · Image generation › Super-resolution · approach: Blind super-resolution · 2 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
Replace the two-branch architecture with a one-branch SRResNetDAN and DASR blind SR · Blur and blur+noisePSNR (dB): DAN blur: 26.168±0.009 -> 26.182±0.011; DAN blur+noise: 27.341±0.072 -> 27.288±0.027; DASR blur: 27.518±0.034 -> 27.573±No clear effect
Use FAIG rather than IG to select the filters to mask (ablation)SRResNet blind SR · Set14 blurry and noisy imagesImage-gradient MSE performance drop (10^-3): Blurry, 1%: 4.31±1.54 -> 6.68±0.63; noisy, 1%: 4.22±0.44 -> 6.62±0.54; blurry, 5%: 6.41±0.88 -> 7.53±0.24; noisy, 5%: 8.Improved
Use FAIG rather than absolute filter changes to select the filters to mask (ablation)SRResNet blind SR · Set14 blurry and noisy imagesImage-gradient MSE performance drop (10^-3): Blurry, 1%: 0.18±0.13 -> 6.68±0.63; noisy, 1%: 0.49±0.10 -> 6.62±0.54; blurry, 5%: 2.16±0.61 -> 7.53±0.24; noisy, 5%: 3.Improved
Use FAIG rather than random selection to choose filters to mask (ablation)SRResNet blind SR · Set14 blurry and noisy imagesImage-gradient MSE performance drop (10^-3): Blurry, 1%: 0.07±0.01 -> 6.68±0.63; noisy, 1%: 0.04±0.01 -> 6.62±0.54; blurry, 5%: 0.55±0.32 -> 7.53±0.24; noisy, 5%: 0.Improved
Retrain only FAIG-selected filters at their selected locations rather than randomly selected filters (ablation)Baseline SRResNet · Set14 blurry images, train for deblurringPSNR (dB): 27.028 -> 28.047Improved
Retrain only FAIG-selected filters at their selected locations rather than randomly selected filters (ablation)Baseline SRResNet · Set14 noisy images, train for denoisingPSNR (dB): 25.526 -> 25.793Improved
Apply deblurring-selected filters to denoising rather than retraining denoising-selected filters (ablation)Baseline SRResNet · Set14 blurry images, train for deblurringPSNR (dB): 27.463 -> 28.047Hurt
Apply denoising-selected filters to deblurring rather than retraining deblurring-selected filters (ablation)Baseline SRResNet · Set14 noisy images, train for denoisingPSNR (dB): 25.590 -> 25.793Hurt
Remove gradient elimination for other degradations when selecting deblurring filters (ablation)FAIG filter selection · Noisy inputs with 1% deblurring filters maskedDenoising function retainedHurt
Use overlap of discovered filter sets to predict degradation without supervised degradation labelsFAIG degradation prediction · DIV2KPrediction accuracy: 98% noise; 96% blurImproved

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