Research map / Image generation / Image restoration
MMDIR: Multimodal Instruction-Driven Framework for Mixed-Degradation Document Image Restoration
Heng Li, Xingyuan Wang, Yang Fan, Yunan Zhang, Xiangping Wu, Qingcai 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 LLM-based text instruction and degradation identification (ablation) | MMDIR · BMVC deblurring | PSNR: 26.76 -> 29.03 | Improved |
| add LLM-based text instruction and degradation identification (ablation) | MMDIR · OSR deshadow | PSNR: 18.79 -> 19.60 | Improved |
| add LLM-based text instruction and degradation identification (ablation) | MMDIR · MixedDoc | PSNR: 23.83 -> 24.43 | Improved |
| add text instruction (ablation) | MMDIR · MixedDoc | LPIPS: 0.1324 -> 0.1217 | Improved |
| add text instruction (ablation) | MMDIR · BMVC deblurring | DISTS: 0.0320 -> 0.0233 | Improved |
| use MMDIR instead of retrained DocDiff and DiffUIR on mixed degradations | MMDIR · MixedDoc | PSNR: 22.09 (DiffUIR1024*) -> 24.43 | Improved |
| use MMDIR instead of retrained FoundIR on mixed degradations | MMDIR · MixedDoc | FID: 10.41 -> 9.07 | Improved |
| use MMDIR for downstream OCR on restored images | MMDIR · MixedDoc | CER: 0.23 (FoundIR*) -> 0.17 | Improved |
| use MMDIR vs LGA-Doc on deblurring SSIM | MMDIR · BMVC deblurring | SSIM: 0.978 (LGA-Doc) vs 0.977 | No clear effect |
| use MMDIR vs BGSNet and DiffUIR on deshadow PSNR | MMDIR · OSR deshadow | PSNR: 21.25 (BGSNet) vs 19.60 | Hurt |
| increase number of co-occurring degradation types from one to four | MMDIR · MixedDoc | PSNR and SSIM | Hurt |
| increase number of degradation types per image | MMDIR degradation identification · MixedDoc | recall: 1 (1 type) -> .9647 (4 types) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FoundIR* — FoundIR: Unleashing Million-scale Training Data to Advance Foundation Models for Image Restoration (ICCV 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Improving Image Restoration through Removing Degradations in Textual Representations (CVPR 2024 · similarity 0.89)
- DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks (CVPR 2024 · similarity 0.88)
- Adapting Text-to-Image Generation with Feature Difference Instruction for Generic Image Restoration (CVPR 2025 · similarity 0.87)
- Learning Where and What to Restore for Composite Image Restoration (NeurIPS 2026 · similarity 0.87)
- DocNLC: A Document Image Enhancement Framework with Normalized and Latent Contrastive Representation for Multiple Degradations (AAAI 2024 · similarity 0.87)
- Descanning: From Scanned to the Original Images with a Color Correction Diffusion Model (AAAI 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.
