SCUT-EnsText

August 1, 2026 · View on GitHub

The SCUT-EnsText Dataset for the research of scene text removal is released by Deep Leaning and Visual Computing Lab of South China University of Technology. The dataset can be downloaded through the following link:

train set - Baidu Cloud (Password : 5xwi) - Google Drive

test set - Baidu Cloud (Password : 8vpg) - Google Drive

The code for EraseNet can be referred to EraseNet.

The SCUT-EnsText dataset can only be used for non-commercial research purpose. Trainging set and testing set are available now, but the training set is encrypted with additional code. To request access, please follow these steps:

Step 1: Download and complete the agreement document:

Have this document signed and stamped by your institution. Please also prepare 1–2 recent publications (within the last 6 years) as evidence that you or your team conduct research in OCR, image inpainting, text editting, and so on.

Step 2: Submit your application online:

🔗 SCUT DLVC Lab Dataset Access Portal → Apply for SCUT-EnsText

Upload both signed documents through the portal and fill out the "Recent Publications" block. Your application will be reviewed manually and you will be notified by email once a decision has been made (typically within 1–5 business days).

Step 3: Download the dataset:

After approval, you will receive the download link and decompression password via email.

⚠️ All users must comply with the use conditions at all times; failure to do so will result in revocation of access.

Dataset Description

The SCUT-EnsText benchmark aims to motivate more advanced deep learning models for scene text removal task. All of the images in our dataset are collected from several public real-world scene text reading benchmarks, including ICDAR2013, ICDAR-2015, MS COCO-Text, SVT, MLT-2017, MLT-2019, and ArTs.

SCUT-EnsText contains a total of 3,562 images with diverse text characteristics, including text shape (horizontal text, arbitrary quadrilateral text and curved text) and languages(English and Chinese). It is split into a training set and a testing set. To ensure that both of them have the same data distribution, we randomly select approximately 70% of the images for training and the remainder of the images for testing. In total, the training set contains 2,749 images with 16,460 words, while the testing set contains 813 images with 4,864 words.

image

Citation and Contact

Please consider to cite our paper when you use our dataset:

@ARTICLE{Erase,
  author={Liu, Chongyu and Liu, Yuliang and Jin, lianwen and Zhang, Shuaitao and Luo, Canjie and Wang, Yongpan},
  journal={IEEE Transactions on Image Processing}, 
  title={EraseNet: End-to-End Text Removal in the Wild}, 
  year={2020},
  volume={29},
  pages={8760-8775},}

For any quetions about the dataset please contact the authors by sending email to Chongyu Liu(liuchongyu1996@gmail.com) or Prof. Jin(eelwjin@scut.edu.cn).

Copyright © 2020 SCUT-DLVC. All Rights Reserved.

Sample