/Awesome-Scene-Text-Image-Super-Resolution

A collection of papers and resources on scene text image super-resolution.

Awesome Scene Text Image Super-Resolution

A comprehensive repository of awesome Scene Text Image Super-Resolution (STISR) methods based on deep learning.

Papers

Year Venue Title Repo Dataset
2023 TCSVT Text-Enhanced Scene Image Super-Resolution via Stroke Mask and Orthogonal Attention TextZoom
2023 AAAI Improving Scene Text Image Super-Resolution via Dual Prior Modulation Network Code TextZoom
2023 TIP Text prior guided scene text image super-resolution Code TextZoom
2023 arXiv Scene Text Image Super-Resolution via Content Perceptual Loss and Criss-Cross Transformer Blocks* TextZoom
2022 - Scene Text Image Super-Resolution with Self-Supervised Memory Learning Code TextZoom
2022 IJCAI C3-STISR: Scene Text Image Super-resolution with Triple Clues Code TextZoom
2022 CVPR A Text Attention Network for Spatial Deformation Robust Scene Text Image Super-resolution* Code TextZoom
2022 AAAI Text gestalt: Stroke-aware scene text image super-resolution Code TextZoom
2021 BMVC Skeleton-aware text image super-resolution TextZoom
2021 Neurocomputing TSRGAN: Real-world text image super-resolution based on adversarial learning and triplet attention TextZoom
2021 ACM MM Scene text image super-resolution via parallelly contextual attention network Code TextZoom
2021 CVPR Scene text telescope: Text-focused scene image super-resolution* Code TextZoom
2021 arXiv Text prior guided scene text image super-resolution@TIP Code TextZoom
2021 ICME A Novel Attention Enhanced Residual-In-Residual Dense Network for Text Image Super-Resolution ICDAR2015-TextSR
2020 ECCV Scene Text Image Super-Resolution in the Wild Code TextZoom
2020 ICASSP Adversarial Text Image Super-Resolution using Sinkhorn Distance ICDAR2015-TextSR
2020 ICPR On-Device Text Image Super Resolution Synth90k
2019 arXiv TextSR: Content-Aware Text Super-Resolution Guided by Recognition Code Synth90k; SynthText
2019 ICME Text-Attentional Conditional Generative Adversarial Network for Super-Resolution of Text Images ICDAR2015-TextSR
  • The "@¥" means that the paper has been Accepted by ¥.
  • The "*" means that the work involves Vision Transformer for STISR.

Quantitative Results

Method Year Venue Accuracy of ASTER
easy medium hard weighted-avg
TSRN 2020 ECCV 75.1% 56.3% 40.1% 58.3%
TEAN 2023 TCSVT 80.4% 64.5% 45.6% 64.6%
  • Accuracy of different methods using ASTER in the Textzoom dataset.

Datasets

Year Venue Title Type Link
2020 ECCV TextZoom Real Dataset
2016 CVPR SynthText Syntic Dataset
2015 ICDAR ICDAR2015-TextSR Syntic Dataset
2014 NIPSW Synth90k Syntic Dataset
  • The use of datasets Synth90K and SynthText for scene text image super-resolution requires manual creation.

Acknowledgment

  • This repository is scheduled to be updated regularyly in accordance with schedules of major AI or CV related Conferences and Journals.

  • Welcome to post Pull-requests to update the list above.

  • Thanks to all the above researchers' contribution.