Research map / Image generation / Autoregressive and latent generation
FreeAdapt: Unleashing Diffusion Priors for Ultra-High-Definition Image Restoration
Xiaoan Liu, Xinyi Liu, Yongjun Zhang, Yi Wan, Li Tingyun, Dongdong Yue
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 |
|---|---|---|---|
| add frequency guidance (FreqG) (ablation) | LDM with patch-based inference · UHD-LL | PSNR: 18.91 -> 21.76 | Improved |
| add feature guidance (FeatG) on top of FreqG (ablation) | LDM FreeAdapt · UHD-LL | LPIPS: 0.314 -> 0.283 | Improved |
| add VAE decoder fine-tuning with skip connection (ablation) | LDM FreeAdapt · UHD-LL | PSNR: 21.88 -> 22.21 | Improved |
| add VAE decoder fine-tuning (ablation) | LDM FreeAdapt · UHD-LL | LPIPS: 0.283 -> 0.253 | Improved |
| add VAE decoder fine-tuning (ablation) | LDM FreeAdapt · UHD-Haze | MUSIQ: 44.75 -> 44.38 | Hurt |
| use phase-only fusion instead of skip residual fusion (ablation) | LDM FreqG · UHD-LL | DISTS: 0.312 -> 0.121 | Improved |
| use phase-only fusion instead of FFT spectrum fusion (ablation) | LDM FreqG · UHD-LL | DISTS: 0.187 -> 0.121 | Improved |
| apply FFSG instead of patch-based inference | DiffBIR · UHD-LL | PSNR: 21.61 -> 23.99 | Improved |
| apply FFSG instead of MultiDiffusion/DemoFusion/PixelSmith | LDM · UHD-LL | PSNR: 21.74 -> 22.21 | Improved |
| add FFSG gain on LDM | LDM · UHD-Blur | LPIPS: 0.213 -> 0.166 | Improved |
| use skip connection in VAE-FT decoder (ablation) | VAE-FT · UHD | visual detail | Improved |
Closest papers by meaning
- Latent Harmony: Synergistic Unified UHD Image Restoration via Latent Space Regularization and Controllable Refinement (NeurIPS 2025 · similarity 0.89)
- RAP-SR: RestorAtion Prior Enhancement in Diffusion Models for Realistic Image Super-Resolution (AAAI 2025 · similarity 0.89)
- Learning Diffusion Texture Priors for Image Restoration (CVPR 2024 · similarity 0.89)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.89)
- FreeControl: Efficient, Training-Free Structural Control via One-Step Attention Extraction (NeurIPS 2025 · similarity 0.89)
- From Zero to Detail: Deconstructing Ultra-High-Definition Image Restoration from Progressive Spectral Perspective (CVPR 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.
