Research map / Image generation / Image restoration
Exploiting Deep Generative Prior for Versatile Image Restoration and Manipulation
Xingang Pan, Xiaohang Zhan, Bo Dai, Dahua Lin, Chen Change Loy, Ping Luo
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use discriminator feature matching loss instead of MSE (ablation) | DGP · ImageNet val 1k (colorization) | ResNet50 accuracy (%): 49.1 -> 56.8 | Improved |
| use discriminator feature matching loss instead of perceptual loss (ablation) | DGP · ImageNet val 1k (colorization) | ResNet50 accuracy (%): 53.9 -> 56.8 | Improved |
| add progressive reconstruction to discriminator loss (ablation) | DGP · ImageNet val 1k (colorization) | ResNet50 accuracy (%): 56.8 -> 62.8 | Improved |
| use discriminator loss instead of MSE (ablation) | DGP · ImageNet val 1k (super-resolution x4) | NIQE: 6.54 -> 6.06 | Improved |
| use discriminator loss instead of perceptual loss (ablation) | DGP · ImageNet val 1k (super-resolution x4) | NIQE: 6.27 -> 6.06 | Improved |
| add progressive reconstruction to discriminator loss (ablation) | DGP · ImageNet val 1k (super-resolution x4) | NIQE: 6.06 -> 4.90 | Improved |
| use discriminator loss instead of MSE (ablation) | DGP · ImageNet val 1k (super-resolution x4) | PSNR: 21.24 -> 21.58 | Improved |
| add progressive reconstruction to discriminator loss (ablation) | DGP · ImageNet val 1k (super-resolution x4) | PSNR: 21.58 -> 22.00 | Improved |
| bias final fine-tuning toward MSE loss instead of discriminator loss (ablation) | DGP · ImageNet val 1k (super-resolution x4) | PSNR and NIQE: PSNR 22.00 -> 23.30, NIQE 4.90 -> 5.30 | Mixed |
| fine-tune the generator on the fly instead of fixing it (GAN inversion) | BigGAN reconstruction · ImageNet val 1k (GAN inversion reconstruction) | PSNR: 22.49 -> 32.89 | Improved |
| use DGP instead of DIP for inpainting | inpainting · ImageNet val 1k (inpainting) | PSNR: 14.58 -> 16.97 | Improved |
| use DGP instead of DIP for adversarial defense | adversarial defense · ImageNet val 1k (adversarial defense) | top-1 accuracy (%): 37.5 -> 41.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SinGAN — SinGAN: Learning a Generative Model From a Single Natural Image (ICCV 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Distilling Semantic Priors from SAM to Efficient Image Restoration Models (CVPR 2024 · similarity 0.90)
- Navigating Image Restoration with VAR's Distribution Alignment Prior (CVPR 2025 · similarity 0.87)
- Low-Light Image Enhancement via Generative Perceptual Priors (AAAI 2025 · similarity 0.87)
- UniRestore: Unified Perceptual and Task-Oriented Image Restoration Model Using Diffusion Prior (CVPR 2025 · similarity 0.87)
- Restoration by Generation with Constrained Priors (CVPR 2024 · similarity 0.85)
- Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild (CVPR 2024 · similarity 0.84)
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.
