🌍 Multilingual OCR Demo

Interactive demo for ScriptMoE, an all-in-one multilingual scene text recognition model based on a script-aware sparse Mixture-of-Experts architecture.

Paper

All-in-One Multilingual Scene Text Recognition with Script-aware Mixture-of-Experts

Xingsong Ye, Yongkun Du, Jiaxin Zhang, Zhixian Li, Chong Sun, Chen Li, Jing Lyu, Lianwen Jin, Zhineng Chen

• 📄 arXiv: https://arxiv.org/abs/2609.24058
• 🤗 Hugging Face Paper: https://huggingface.co/papers/2609.24058
• 💻 GitHub: https://github.com/YesianRohn/ScriptMoE

About ScriptMoE

ScriptMoE is an all-in-one multilingual scene text recognizer designed to handle diverse writing systems with a single model.

The model uses:

• a shared visual encoder
• an image-level script-aware router
• top-2 script-aligned experts
• a shared expert for cross-script knowledge
• a sparse Mixture-of-Experts decoder

The model is trained with multilingual supervision from TextMuSS-10M, covering 10 scripts and 229 languages.

Try the Demo

Upload an image containing scene text and run the model to obtain multilingual text recognition results.

The demo is intended to provide a quick and interactive way to explore the multilingual recognition capability of ScriptMoE.

Resources

Results

In the paper, ScriptMoE achieves:

• 82.06% average word accuracy on TextMuSS-Bench
• 80.89% F1 on the CC-OCR multilingual end-to-end task when replacing the recognizer in PP-OCRv5

TextMuSS-Bench contains 10 scripts and 10,899 real-world images.

Citation

If you use this demo or ScriptMoE in your research, please cite:

@article{ye2026scriptmoe,
  title   = {All-in-One Multilingual Scene Text Recognition with Script-aware Mixture-of-Experts},
  author  = {Ye, Xingsong and Du, Yongkun and Zhang, Jiaxin and Li, Zhixian and Sun, Chong and Li, Chen and Lyu, Jing and Jin, Lianwen and Chen, Zhineng},
  journal = {arXiv preprint arXiv:2609.24058},
  year    = {2026}
}

arXiv:2609.24058

https://arxiv.org/abs/2609.24058