Research map / Image generation / Super-resolution
Enhanced Generative Structure Prior for Chinese Text Image Super-Resolution
Xiaoming Li, Wangmeng Zuo, Chen Change Loy
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 |
|---|---|---|---|
| upgrade to MARCONet++ with per-character style w and irregular-layout prior | MARCONet · Synthetic Chinese irregular x4 | PSNR: 21.22 -> 23.86 | Improved |
| add StyleGAN generative prior (ablation) | MARCONet++ (w/o S) · Synthetic irregular x4 | PSNR: 21.45 -> 23.86 | Improved |
| add codebook of character codes (ablation) | MARCONet++ (w/o C) · Synthetic irregular x4 | PSNR: 21.20 -> 23.86 | Improved |
| enlarge UNet capacity without structure prior (ablation) | UNet · Synthetic irregular x4 | PSNR: 19.68 -> 19.89 | No clear effect |
| concatenate structure image to discriminator input (ablation) | MARCONet++ · Synthetic irregular x4 | PSNR: 23.73 -> 23.86 | Improved |
| use multi-scale prior transform instead of 32x32 only (ablation) | MARCONet++ · Synthetic irregular x4 | PSNR: 23.66 -> 23.86 | Improved |
| use multi-scale prior transform instead of 64x64 only (ablation) | MARCONet++ · Synthetic irregular x4 | PSNR: 23.81 -> 23.86 | Improved |
| add adjacent-style regularization lambda_reg 0.02 (ablation) | MARCONet++ · Synthetic irregular x4 | PSNR: 23.58 -> 23.86 | Improved |
| raise lambda_reg from 0.02 to 0.1 (ablation) | MARCONet++ · Synthetic irregular x4 | PSNR: 23.86 -> 23.61 | Hurt |
| add perceptual loss lambda_percep 0.05 (ablation) | MARCONet++ · Synthetic irregular x4 | LPIPS: .197 -> .125 | Improved |
| raise text length from 8 to 32 characters (ablation) | MARCONet++ · Synthetic 32-char text crops | character accuracy: 87.1 -> 48.1 | Hurt |
| train location and recognition branches separately | MARCONet++ · Synthetic Chinese | accuracy: 12.0% ACC, 41.2% LOC | Improved |
Closest papers by meaning
- VQ-FONT: Few-Shot Font Generation with Structure-Aware Enhancement and Quantization (AAAI 2024 · similarity 0.88)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.87)
- PRISM: Prior Rectification and Uncertainty-Aware Structure Modeling for Diffusion-Based Text Image Super-Resolution (NeurIPS 2026 · similarity 0.87)
- GlyphMastero: A Glyph Encoder for High-Fidelity Scene Text Editing (CVPR 2025 · similarity 0.86)
- DeepCalliFont: Few-Shot Chinese Calligraphy Font Synthesis by Integrating Dual-Modality Generative Models (AAAI 2024 · similarity 0.85)
- AnyText: Multilingual Visual Text Generation and Editing (ICLR 2024 · similarity 0.85)
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.
