AI Paper Atlas

Research map / Image generation / Image restoration

On Measuring and Controlling the Spectral Bias of the Deep Image Prior

Zenglin Shi, Pascal Mettes, Subhransu Maji, Cees G. M. Snoek

IJCV 2022 · Image generation › Image restoration · approach: Priors for restoration networks · 88 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
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · CBSD68 denoising sigma=15PSNR: 30.58 -> 30.80Improved
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · CBSD68 denoising sigma=50PSNR: 24.59 -> 24.83Improved
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · LIVE1 JPEG deblocking quality=10PSNR: 27.52 -> 27.70Improved
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · CBSD68 region inpainting ratio=0.5PSNR: 17.26 -> 17.36Improved
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · Set14 4x super-resolutionPSNR: 27.00 -> 27.50Improved
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · Set14 8x super-resolutionPSNR: 24.15 -> 24.17No clear effect
use controlled spectral bias instead of DIP with oracle stoppingDecoder with Lipschitz normalization and Gaussian upsampling · Set5 4x super-resolutionPSNR: 29.89 -> 30.45Improved
compare unsupervised method with supervised networksDecoder with Lipschitz normalization and Gaussian upsampling · CBSD68 denoising sigma=25PSNR: 28.15 vs 31.21 (FFDNet)Hurt
raise noise level to sigma=100Decoder with Lipschitz normalization and Gaussian upsampling · image denoising F16qualitative: sigma=25 works, sigma=100 failsHurt
decrease lambda in controlled priorDecoder with Lipschitz normalization and Gaussian upsampling · image detail enhancementsmoothness: smaller lambda gives smoother outputImproved

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