AI Paper Atlas

Research map / Image generation / Super-resolution

Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance

Minxing Luo, Linlong Fan, Qiushi Wang, Ge Wu, Yiyan Luo, Yuhang Yu, Jinwei Chen, Yaxing Wang, Qingnan Fan, Jian Yang

CVPR 2026 · Image generation › Super-resolution · approach: Perceptual and real-world SR

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 changedWhereMeasuredEffect
guide stage 2 with the restored glyph mask instead of an empty mask (ablation)TiGeSR · Real-CE (x4)OCR-A: 59.5\% -> 67.3\%Improved
use restored glyph mask instead of standard-font rendered mask (ablation)TiGeSR · Real-CE (x4)OCR-A: 55.3\% -> 67.3\%Improved
use the text restoration pipeline instead of SAM-TS extraction (ablation)TiGeSR · Real-CE (x4)OCR-A: 57.9\% -> 67.3\%Improved
use two-phase RGB+mask training instead of a plain latent diffusion mask model (ablation)TiGeSR · Real-CE (x4)OCR-A: 60.1\% -> 67.3\%Improved
provide OCR-predicted text in stage 1 instead of null text (ablation)TiGeSR · UZ-ST 35mm subsetOCR-A: 40.4\% -> 44.6\%Improved
replace OCR text with random text (ablation)TiGeSR · UZ-ST 35mm subsetOCR-A: 44.6\% -> 40.3\%Hurt
add UZ-ST to training (ablation)TiGeSR · UZ-ST (all zoom levels, average)OCR-A: 40.0\% -> 43.0\%Improved
add UZ-ST to fine-tuning of OSEDiff (ablation)OSEDiff · UZ-ST (all zoom levels, average)OCR-A: 22.8\% -> 28.9\%Improved
add UZ-ST to fine-tuning of DiT4SR (ablation)DiT4SR · UZ-ST (all zoom levels, average)OCR-A: 19.3\% -> 23.7\%Improved
add UZ-ST to fine-tuning of OSEDiff (ablation)OSEDiff · UZ-ST (all zoom levels, average)LPIPS: 0.197 -> 0.201Hurt
two-stage text-first design vs strongest one-stage baselinesTiGeSR · Real-CE (x4) text cropsdelta OCR-A vs LR: -1.0\% (TADiSR) -> +2.5\%Improved

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy