Research map / Image generation / Text-to-image diffusion
Text-Aware Image Restoration with Diffusion Models
Jaewon Min, Jin Hyeon Kim, Paul Hyunbin Cho, Jaeeun Lee, Jihye Park, Park Min Kyu, Sangpil Kim, Hyunhee Park, Seungryong Kim
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 text prompts from the spotter at inference (null -> predicted) (ablation) | TeReDiff stage 3 · SA-Text Level2 | end-to-end recognition (None): 23.46 -> 26.39 | Improved |
| use ground-truth text prompt instead of predicted (ablation) | TeReDiff stage 3 · SA-Text Level2 | end-to-end recognition (None): 26.39 -> 33.31 | Improved |
| advance training from stage 1 to joint stage 3 (ablation) | TeReDiff · SA-Text Level2 | end-to-end recognition (None): 21.24 -> 23.46 (null prompt) | Improved |
| use diffusion features instead of ResNet-50 for text spotting (100K) (ablation) | TESTR · SA-Text test | end-to-end recognition (None): 17.03 -> 32.29 | Improved |
| use diffusion features instead of ResNet-50 with 20K data (ablation) | TESTR · SA-Text test | end-to-end recognition (None): 1.81 -> 27.42 | Improved |
| caption-style prompt instead of tag-style with predicted text (ablation) | TeReDiff stage 3 · SA-Text Level2 | end-to-end recognition (None): 24.26 -> 26.39 (recall 59.25 -> 52.86) | Mixed |
| caption vs tag prompt style with GT text (ablation) | TeReDiff stage 3 · SA-Text Level2 | end-to-end recognition (None): 32.02 -> 33.31 | No clear effect |
| use TeReDiff instead of prior diffusion restorers | TeReDiff · SA-Text Level3 | text detection F1: 50.65 -> 56.13 (vs DiffBIR) | Improved |
| apply prior restorers to strongly degraded images | SeeSR / DiffBIR etc. · SA-Text Level2/3 | end-to-end recognition: LQ 20.49 vs DiffBIR 21.64; Real-Text LQ 47.08 vs DiffBIR 39.27 | Hurt |
| use TeReDiff vs DiffBIR on image quality | TeReDiff · SA-Text test | non-reference metrics (MANIQA): LPIPS 0.3636 -> 0.2828; MANIQA 0.6771 -> 0.6471 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FaithDiff — FaithDiff: Unleashing Diffusion Priors for Faithful Image Super-resolution (CVPR 2025) · better on 6 of 6 shared results
- SeeSR — SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024) · better on 4 of 4 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
- Diffusion-based Blind Text Image Super-Resolution (CVPR 2024 · similarity 0.89)
- Improving Image Restoration through Removing Degradations in Textual Representations (CVPR 2024 · similarity 0.89)
- Text-Aware Real-World Image Super-Resolution via Diffusion Model with Joint Segmentation Decoders (NeurIPS 2025 · similarity 0.89)
- One Stone with Two Birds: A Null-Text-Null Frequency-Aware Diffusion Models for Text-Guided Image Inpainting (NeurIPS 2025 · similarity 0.89)
- Adapting Text-to-Image Generation with Feature Difference Instruction for Generic Image Restoration (CVPR 2025 · similarity 0.88)
- Enhancing Text-to-Image Diffusion Transformer via Split-Text Conditioning (NeurIPS 2025 · similarity 0.88)
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.
