Research map / Image generation / Image restoration
LIRA: Lifelong Image Restoration from Unknown Blended Distortions
Jianzhao Liu, Jianxin Lin, Xin Li, Wei Zhou, Sen Liu, Zhibo Chen
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 pre-trained experts (ablation) | LIRA base network · DIV2K moderate | PSNR: 25.95 -> 27.24 | Improved |
| add SK-Blocks in place of DuRB-M (ablation) | LIRA base network · DIV2K moderate | PSNR: 26.76 -> 27.24 | Improved |
| add gate layers (all) (ablation) | LIRA base network · DIV2K moderate | PSNR: 26.64 -> 27.24 | Improved |
| add second half of gate layers (ablation) | LIRA base network · DIV2K moderate | PSNR: 27.14 -> 27.24 | Improved |
| add attention layer for expert mixing (ablation) | LIRA base network · DIV2K moderate | PSNR: 26.74 -> 27.24 | Improved |
| add Bi-SGUs instead of concat conv (ablation) | LIRA base network · DIV2K moderate | PSNR: 27.18 -> 27.24 | Improved |
| replay GAN pseudo old samples with base network responses | Setting A (new data only) · Haze, moderate old task | PSNR: 26.50 -> 27.24 | Improved |
| lower learning rate of old expert parameters (ablation) | Fine-tuning · Haze, mild old task | PSNR: 20.25 -> 28.03 | Improved |
| use pseudo old samples instead of real old samples | Joint learning · Haze, moderate old task | PSNR: 27.25 -> 27.24 | No clear effect |
| use LIRA instead of EWC regularization | EWC · Haze, moderate old task | PSNR: 20.86 -> 27.24 | Improved |
| train expanded branch instead of freezing old expert | PNN · Haze, moderate new task | PSNR: 24.21 -> 26.05 | Improved |
| use LIRA base network instead of OWAN | OWAN · BSDS100 128x128 | PSNR: 23.70 -> 25.24 | Improved |
Closest papers by meaning
- Beyond the Static-World: Lifelong Learning for All-in-One Medical Image Restoration (CVPR 2026 · similarity 0.87)
- Expert Weaving: Marrying Masked AutoRegressive and Diffusion Models for Unified Image Restoration (ECCV 2026 · similarity 0.87)
- Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration (NeurIPS 2026 · similarity 0.86)
- Complexity Experts are Task-Discriminative Learners for Any Image Restoration (CVPR 2025 · similarity 0.85)
- Parameter Efficient Adaptation for Image Restoration with Heterogeneous Mixture-of-Experts (NeurIPS 2024 · similarity 0.85)
- A Restoration Network as an Implicit Prior (ICLR 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.
