Home

Awesome

<div align=center>

Revisiting Scene Text Recognition: A Data Perspective

</div> <div align=center> <img src='https://github.com/open-mmlab/mmocr/assets/65173622/4544c4ff-0f30-46b2-ae04-7bd795694df4' width=600 > </div> <div align=center> <p >Union14M is a large scene text recognition (STR) dataset collected from 17 publicly available datasets, which contains 4M of labeled data (Union14M-L) and 10M of unlabeled data (Union14M-U), intended to provide a more profound analysis for the STR community</p> <div align=center>

arXiv preprint Gradio demo Open In Colab

</div> </div> <p align="center"> <strong><a href="#1-introduction">Introduction </a></strong> • <strong><a href="#34-download">Download </a></strong> • <strong><a href="#5-maerec">MAERec</a></strong> </p>

What's New

1. Introduction

<div align=center> <img src='https://github.com/open-mmlab/mmocr/assets/65173622/4ec60ac0-ca4e-4233-a196-8e9e46a0c21d' width=400 > <img src='https://github.com/open-mmlab/mmocr/assets/65173622/2991aa45-01cc-44da-a62f-56fd16012ab2' width=400 > </div>

2. Contents

3. Union14M Dataset

3.1. Union14M-L

<div align=center> <img src='https://github.com/open-mmlab/mmocr/assets/65173622/79c4b5a5-4a2f-46da-aeca-3d92e9199861' width=700 > </div>

3.2. Union14M-U

<div align=center> <img src='https://github.com/open-mmlab/mmocr/assets/65173622/11711617-0fb0-42e0-a6ec-1938e2a71e61' width=600 > </div>

3.3. Union14M-Benchmark

<div align=center> <img src='https://github.com/open-mmlab/mmocr/assets/65173622/bcfdd369-decd-4b6f-903b-3021531bd119' width=600 > </div>

3.4. Download

DatasetsOne DriveBaidu Yun
Union14M-L & Union14M-Benchmark (12GB)One DriveBaidu Yun
Union14M-U (36.63GB)One DriveBaidu Yun
6 Common Benchmarks (17.6MB)One DriveBaidu Yun
<!-- TODO: Add Google Drive Links -->

4. STR Models trained on Union14M-L

4.1. Checkpoints

5. MAERec

5.1. Pre-training

5.2. Fine-tuning

5.3. Evaluation

5.4. Inferencing

5.5. Demo

<div align=center> <img src='https://github.com/open-mmlab/mmocr/assets/65173622/829741e1-ca5c-4be7-9a25-ce14da4f9d50' width=600 > </div>

6. License

7. Acknowledgment

8. Citation

@inproceedings{jiang2023revisiting,
      title={Revisiting Scene Text Recognition: A Data Perspective}, 
      author={Qing Jiang and Jiapeng Wang and Dezhi Peng and Chongyu Liu and Lianwen Jin}
      booktitle={Proceedings of the IEEE/CVF international conference on computer vision},
      year={2023},
}