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.