Research map / Reinforcement learning / Offline RL
Beyond the Boundary: RL-Driven Solution Space Exploration for Blind Face Restoration
Bin WU, Wei Wang, Yahui Liu, Chi Zhang, Yao Zhao
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 |
|---|---|---|---|
| apply LRPO online RL post-training | DiffBIR · CelebA-Test | SSIM: 0.6775 -> 0.7021 | Improved |
| apply LRPO online RL post-training | DiffBIR · CelebA-Test | Deg.: 35.16 -> 31.19 | Improved |
| apply LRPO online RL post-training | DiffBIR · WebPhoto-Test | MUSIQ: 67.45 -> 72.71 | Improved |
| apply LRPO online RL post-training | DiffBIR · LFW-Test | CLIPIQA: 0.7296 -> 0.8073 | Improved |
| add RL (no KL) to base model (ablation) | DiffBIR · CelebA-Test | CLIPIQA: 0.7200 -> 0.7809 | Improved |
| add KL divergence term to RL objective (ablation) | Variant 1 · CelebA-Test | SSIM: 0.6849 -> 0.6750 | Hurt |
| add noise-level advantage assignment (AdA) (ablation) | LRPO w/o AdA (Variant 3) · CelebA-Test | SSIM: 0.6867 -> 0.7021 | Improved |
| add GT-guided likelihood regularization (ablation) | LRPO w/o Reg (Variant 4) · CelebA-Test | SSIM: 0.6806 -> 0.7021 | Improved |
| add RL exploration to GT-guided regularization (ablation) | Reg only · CelebA-Test | CLIPIQA: 0.3654 -> 0.8061 | Improved |
| use RL group exploration instead of same-reward SFT (ablation) | LRPO · CelebA-Test | Deg.: 36.35 -> 31.19 | Improved |
| add perceptual reward r_aq and r_fid to r_pref reward (ablation) | composite reward · CelebA-Test | Deg.: 36.93 -> 31.19 | Improved |
| add r_aq to r_pref + r_fid reward (ablation) | composite reward · CelebA-Test | CLIPIQA: 0.7355 -> 0.8061 | Improved |
Closest papers by meaning
- $InterLCM$: Low-Quality Images as Intermediate States of Latent Consistency Models for Effective Blind Face Restoration (ICLR 2025 · similarity 0.88)
- Overcoming False Illusions in Real-World Face Restoration with Multi-Modal Guided Diffusion Model (ICLR 2025 · similarity 0.86)
- Self-Supervised Selective-Guided Diffusion Model for Old-Photo Face Restoration (NeurIPS 2025 · similarity 0.86)
- OSDFace: One-Step Diffusion Model for Face Restoration (CVPR 2025 · similarity 0.86)
- Blind Face Restoration under Extreme Conditions: Leveraging 3D-2D Prior Fusion for Superior Structural and Texture Recovery (AAAI 2024 · similarity 0.86)
- ReF-LDM: A Latent Diffusion Model for Reference-based Face Image Restoration (NeurIPS 2024 · similarity 0.86)
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.
