AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

What Is It Like to Be a Noise? An Entropy-based Gaussian Noise Regularization for Diffusion Models

Pascal Chang, Kai Lascheit, Jingwei Tang, Markus Gross, Vinicius C. Azevedo

CVPR 2026 · Image generation › Fast diffusion sampling · approach: Denoising diffusion probabilistic models

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 2D spatial entropy term to 1D value entropy (ablation)Gaussianity regularizer · checkerboard latent ablation (qualitative)noise qualityImproved
add Bethe correction term (ablation)Gaussianity regularizer · checkerboard latent ablation (qualitative)statistical convergence at low-resolution scalesImproved
add multilevel (multi-scale) loss (ablation)Gaussianity regularizer · checkerboard latent ablation (qualitative)noise qualityImproved
add regularizer vs no regularization in aesthetic reward optimizationSDXL-Turbo noise optimization · Aesthetic score rewardImageReward: -0.950 -> 0.826Improved
add regularizer vs no regularization in aesthetic reward optimizationSDXL-Turbo noise optimization · Aesthetic score rewardAesthetic score: 7.975 -> 6.478Hurt
our regularizer vs Hwang et al. on aesthetic rewardSDXL-Turbo noise optimization · Aesthetic score rewardCLIP score: 25.019 vs 24.574No clear effect
our regularizer vs Hwang et al. on brightness minimizationSDXL-Turbo noise optimization · Brightness minimizationImageReward: 0.925 -> 1.038Improved
add regularizer vs no regularization in brightness minimizationSDXL-Turbo noise optimization · Brightness minimizationCLIP score: 18.606 -> 27.914Improved
use KDE entropy with 128 fixed binsGaussianity regularizer · compute costtime/step (ms): 0.7148 (Hwang et al.) vs 15.0191 (Ours)Hurt
optimize noise for Pearson correlation with target image using regularizerSD1.5, SD2.1, SD-Turbo · model-free image-to-noise matchinglayout similarity (qualitative)Improved
optimize from a clean latent at low learning rateGaussianity regularizer · failure caseoutput quality (qualitative)Hurt

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