Research map / Image generation / Super-resolution
SPIRE: Semantic Prompt-Driven Image Restoration
Chenyang QI, Zhengzhong Tu, Keren Ye, Mauricio Delbracio, Peyman Milanfar, Qifeng Chen, Hossein Talebi
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 semantic prompt and restoration prompt to ControlNet-SR (ablation) | ControlNet-SR · MS-COCO parameterized degradation | FID: 13.65 -> 12.14 | Improved |
| add modulation fusion layer on skip features (ablation) | ControlNet with prompts · MS-COCO parameterized degradation | FID: 12.14 -> 11.34 | Improved |
| also modulate backbone upsampling features f_up (ablation) | modulation fusion layer · MS-COCO parameterized degradation | FID: 11.34 -> 11.21 | No clear effect |
| embed degradation parameters (e.g. sigma) in restoration prompt (ablation) | SPIRE · MS-COCO parameterized degradation | FID: 11.58 -> 11.34 | Improved |
| add semantic prompt to the image-to-image model (ablation) | SPIRE baseline with modulation layer · MS-COCO parameterized degradation | FID: 13.60 -> 11.71 | Mixed |
| zero-shot combination of semantic prompt with existing models via CLIP or SDEdit | DiffBIR · MS-COCO parameterized degradation | LPIPS: 0.302 -> 0.365 (DiffBIR vs DiffBIR + CLIP) | Hurt |
| use detailed full semantic prompt instead of short prompt (ablation) | SPIRE · RealPhoto60 | MUSIQ: 61.68 -> 67.82 | Improved |
| use unmatched long semantic prompt instead of short prompt (ablation) | SPIRE · RealPhoto60 | MUSIQ: 61.68 -> 67.54 (short -> unmatched uses 60.27 vs full 67.82 per column order) | Mixed |
| use blind prompt Remove all degradation instead of manual deblur prompt (ablation) | SPIRE · RealPhoto60 | MUSIQ: 35.82 -> 61.68 | Improved |
| increase deblur strength in restoration prompt (ablation) | SPIRE · RealPhoto60 | MUSIQ: 35.82 -> 49.98 (0.6 -> 2.0) | Improved |
| finetune SPIRE on StableSR training set (ablation) | SPIRE without text · DIV2K testset | FID: 28.80 -> 22.45 | Improved |
Closest papers by meaning
- Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild (CVPR 2024 · similarity 0.89)
- Adapting Text-to-Image Generation with Feature Difference Instruction for Generic Image Restoration (CVPR 2025 · similarity 0.89)
- PromptFix: You Prompt and We Fix the Photo (NeurIPS 2024 · similarity 0.89)
- InstructRestore: Region-Customized Image Restoration with Human Instructions (NeurIPS 2025 · similarity 0.88)
- Controlling Vision-Language Models for Multi-Task Image Restoration (ICLR 2024 · similarity 0.88)
- Object-Aware Inversion and Reassembly for Image Editing (ICLR 2024 · similarity 0.87)
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.
