Research map / Image generation / Image restoration
PromptIR: Prompting for All-in-One Image Restoration
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 prompt block (fixed prompts) to Restormer (ablation) | PromptIR · Rain100L | PSNR: 36.74 -> 36.85 | Helped |
| make prompts dynamic and input-conditioned via PGM instead of fixed (ablation) | PromptIR · Rain100L | PSNR: 36.85 -> 37.04 | Helped |
| add prompt block overall to Restormer baseline (ablation) | PromptIR · Rain100L | PSNR: 36.74 -> 37.04 | Helped |
| add prompt blocks at levels 3 and 2 in addition to latent level (ablation) | PromptIR · Rain100L | PSNR: 36.76 -> 37.04 | Helped |
| replace prompt tokens with AirNet contrastive degradation embeddings (ablation) | PromptIR · All-in-one average | PSNR: 32.06 -> 29.24 | Hurt |
| add prompt blocks to encoder as well as decoder (ablation) | PromptIR · All-in-one average | PSNR: 32.06 -> 31.14 | Hurt |
| train on three degradation types instead of noise+rain only | PromptIR · Dehazing/Deraining all-in-one | PSNR: 39.32 -> 36.37 deraining | Mixed |
| add haze data to noise-only training | PromptIR · BSD68 sigma=50 | PSNR: 28.49 -> 28.06 | Hurt |
| use PromptIR instead of AirNet on unseen noise sigma=100 | AirNet · BSD68 | PSNR: 13.64 -> 21.03 | Helped |
| use PromptIR instead of Restormer on single-task dehazing | Restormer · SOTS | PSNR: 30.87 -> 31.31 | Helped |
| use PromptIR instead of AirNet in all-in-one setting | AirNet · Average PSNR | PSNR: 31.20 -> 32.06 | Helped |
Closest papers by meaning
- UP-Restorer: When Unrolling Meets Prompts for Unified Image Restoration (AAAI 2025 · similarity 0.88)
- SelfPromer: Self-Prompt Dehazing Transformers with Depth-Consistency (AAAI 2024 · similarity 0.87)
- DDR: Exploiting Deep Degradation Response as Flexible Image Descriptor (NeurIPS 2024 · similarity 0.87)
- DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation (NeurIPS 2024 · similarity 0.86)
- Controlling Vision-Language Models for Multi-Task Image Restoration (ICLR 2024 · similarity 0.86)
- Dual Prompting Image Restoration with Diffusion Transformers (CVPR 2025 · 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.