Research map / Image generation / Image restoration
Invertible generative models for inverse problems: mitigating representation error and dataset bias
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 latent-norm regularization (gamma) to Glow denoising objective | Glow invertible prior · CelebA denoising | PSNR: outperforms BM3D by almost 2 dB | Helped |
| Replace Glow prior with DCGAN prior in denoising | DCGAN prior · CelebA denoising | PSNR | Hurt |
| Use invertible Glow prior instead of low-dimensional GAN prior for compressive sensing (in-distribution) | Glow prior · CelebA-HQ compressive sensing (in-distribution, 64px and 128px) | PSNR | Helped |
| Use invertible Glow prior instead of GAN prior for out-of-distribution compressive sensing (FFHQ) | Glow prior · FFHQ compressive sensing (out-of-distribution) | PSNR | Mixed |
| Initialize latent optimization at z0=0 without explicit likelihood penalty in compressive sensing (ablation) | Glow prior · CelebA compressive sensing | PSNR | Helped |
| Use a GAN prior with low latent dimensionality vs invertible prior at large measurement counts | DCGAN and PGGAN priors · CelebA compressive sensing | PSNR | Hurt |
| Add pixelwise learned 1x1 convolution and affine coupling (Glow) as invertible prior architecture | Invertible prior · general | qualitative | No clear effect |
Closest papers by meaning
- Algorithmic Guarantees for Inverse Imaging with Untrained Network Priors (NeurIPS 2019 · similarity 0.93)
- pcaGAN: Improving Posterior-Sampling cGANs via Principal Component Regularization (NeurIPS 2024 · similarity 0.90)
- An analytic theory of convolutional neural network inverse problems solvers (ICML 2026 · similarity 0.90)
- A Diffusion Model with State Estimation for Degradation-Blind Inverse Imaging (AAAI 2024 · similarity 0.89)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.89)
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.88)
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.