Research map / Image generation / Fast diffusion sampling
Differentiable Gaussianization Layers for Inverse Problems Regularized by Deep Generative Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add Gaussianization layers (ICA + 1D Gaussianization) to latent reparameterization | StyleGAN2 CS-MRI · CS-MRI Accl=8x SNR=20 dB | PSNR: 27.99 vs 26.93 spherical | Helped |
| Add Gaussianization layers to latent reparameterization | StyleGAN2 CS-MRI · CS-MRI Accl=8x SNR=10 dB | PSNR: 25.48 vs 22.78 spherical | Helped |
| Add Gaussianization layers to latent reparameterization | StyleGAN2 CS-MRI · CS-MRI Accl=2x SNR=20 dB | PSNR: 32.41 vs TV 32.92 | No clear effect |
| Add Gaussianization layers to latent reparameterization | Glow deblurring · CelebA-HQ deblurring | LPIPS: 0.13 vs 0.17 spherical | Helped |
| Add Gaussianization layers to latent reparameterization | StyleGAN2 eikonal tomography · eikonal tomography | PSNR: 24.80 vs 22.19 spherical | Helped |
| Add ICA layer (orthogonal decorrelation of patch entries) to Gaussianization pipeline (ablation) | StyleGAN2 CS-MRI · CS-MRI Accl=8x SNR=20 dB | PSNR: 27.91 vs 26.93 with no components | Helped |
| Add ZCA/whitening layer alone without ICA (ablation) | StyleGAN2 CS-MRI · CS-MRI Accl=8x SNR=20 dB | PSNR: 27.08 vs 26.93 | No clear effect |
| Add Yeo-Johnson 1D Gaussianization on top of ICA (ablation) | StyleGAN2 CS-MRI · CS-MRI Accl=8x SNR=10 dB | PSNR: 25.48 vs 25.22 (ICA only) | Helped |
| Add Lambert W x F_X 1D Gaussianization on top of ICA (ablation) | StyleGAN2 CS-MRI · CS-MRI Accl=8x SNR=10 dB | PSNR: 25.57 vs 25.22 (ICA only) | Helped |
| Add ICA layer to Glow Gaussianization (P2) (ablation) | Glow deblurring · CelebA-HQ deblurring | LPIPS: 0.14 vs 0.17 without components | Helped |
| Add 1D Gaussianization (YJ or Lambert) on Glow beyond ICA (ablation) | Glow deblurring · CelebA-HQ deblurring | PSNR: 22.40 to 22.52 range among ICA-containing variants | No clear effect |
| Increase patch size of style vectors from 8 to 64 (ablation) | StyleGAN2 CS-MRI (ICA+standardization) · CS-MRI Accl=8x SNR=20 dB | PSNR: 28.03 at 64 vs 25.33 at 8 | Helped |
Closest papers by meaning
- pcaGAN: Improving Posterior-Sampling cGANs via Principal Component Regularization (NeurIPS 2024 · similarity 0.87)
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.86)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.85)
- A Unified Framework for Forward and Inverse Problems in Subsurface Imaging using Latent Space Translations (ICLR 2025 · similarity 0.85)
- Gradient-free Decoder Inversion in Latent Diffusion Models (NeurIPS 2024 · similarity 0.85)
- Sparse Bayesian Generative Modeling for Compressive Sensing (NeurIPS 2024 · similarity 0.85)
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.