Research map / Image generation / Super-resolution
Disentangled Textual Priors for Diffusion-based Image Super-Resolution
Lei Jiang, Xin Liu, Xinze Tong, Zhiliang Li, Jie Liu, Jie Tang, Gangshan Wu
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 local (object-level) text priors (ablation) | DTPSR · DRealSR | MUSIQ: 67.48 -> 68.86 | Improved |
| add global text prior on top of local priors (ablation) | DTPSR · DRealSR | MUSIQ: 68.86 -> 69.24 | Improved |
| add global prior only (no local) (ablation) | DTPSR · DRealSR | MUSIQ: 67.48 -> 67.80 | Improved |
| add high-frequency prior on top of low-frequency prior (ablation) | DTPSR · DRealSR | CLIP-IQA: 0.7544 -> 0.7640 | Improved |
| add low-frequency prior on top of high-frequency prior (ablation) | DTPSR · DRealSR | MANIQA: 0.5495 -> 0.6011 | Improved |
| disentangle LF and HF into separate cross-attention branches instead of a mixed sentence (ablation) | DTPSR · DRealSR | CLIP-IQA: 0.7527 -> 0.7640 | Improved |
| add multi-branch negative prompts instead of none (ablation) | DTPSR · DRealSR | MUSIQ: 66.73 -> 69.24 | Improved |
| use single generic negative prompt instead of none (ablation) | DTPSR · DRealSR | MANIQA: 0.5995 -> 0.5776 | Hurt |
| multi-branch negative prompts instead of single generic negative prompt (ablation) | DTPSR · DRealSR | CLIP-IQA: 0.7499 -> 0.7640 | Improved |
| corrupt textual descriptions by replacing words with None (ablation) | DTPSR · Table 7 (labelled DRealSR) | MUSIQ: 71.84 -> 71.64 | No clear effect |
| disentangled textual priors vs best baselines on perceptual metrics | DTPSR · DIV2K-Val | PSNR: 24.62 -> 22.03 | Hurt |
| use DTPSR vs FaithDiff | DTPSR · DRealSR | MUSIQ: 66.30 -> 69.24 | Improved |
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 6 of 6 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
- The Power of Context: How Multimodality Improves Image Super-Resolution (CVPR 2025 · similarity 0.91)
- SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024 · similarity 0.91)
- Pixel-level and Semantic-level Adjustable Super-resolution: A Dual-LoRA Approach (CVPR 2025 · similarity 0.90)
- PRISM: Prior Rectification and Uncertainty-Aware Structure Modeling for Diffusion-Based Text Image Super-Resolution (NeurIPS 2026 · similarity 0.90)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.90)
- Text-Aware Real-World Image Super-Resolution via Diffusion Model with Joint Segmentation Decoders (NeurIPS 2025 · similarity 0.89)
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.
