PaddleOCR 无法正确识别西里尔字母（俄语）— 输出结果为音译 . PaddleOCR Fails to Recognize Cyrillic (Russian) — Output Shows Transliteration

[stormA4ka]

GitHub Issue 模板 (中文版)

标题:
🚨 PaddleOCR 无法正确识别西里尔字母（俄语）— 输出结果为音译

问题描述:
您好！感谢您开发了如此优秀的PaddleOCR工具。🇨🇳🤝🇷🇺

我在使用PaddleOCR识别俄语文本（西里尔字母）时遇到了问题。模型可以运行，但输出的不是西里尔字母，而是拉丁字母的音译结果。

问题复现

输入代码:

from paddleocr import PaddleOCR
ocr = PaddleOCR(lang="ru")  
img_path = "https://book.sibverk.ru/image/trumb/500x500/13-921550.jpg"  
result = ocr.predict(img_path)  
for line in result:  
    print(line['rec_texts'])  

实际输出:

['TMMH POCCNMCKOM 中EAEPALMN']

期望输出:

['ГИМН РОССИЙСКОЙ ФЕДЕРАЦИИ']

已尝试方案

1. 完全按照官方文档安装:

   python -m pip install paddlepaddle-gpu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
   pip install paddleocr==3.0.0

2. 确认已设置 lang="ru" 参数
3. 测试不同图片问题依旧存在

疑问

1. 这是已知问题吗？是否有解决方案？
2. 是否需要额外下载西里尔字母的语言模型？
3. 是否存在后处理参数可以修正音译结果？

环境信息

• 操作系统: （例如 Ubuntu 22.04）
• Python版本: （例如 3.10.12）
• PaddlePaddle: 3.0.0
• PaddleOCR: 3.0.0

衷心感谢您的帮助！中国开发者的技术令人钦佩，我们俄罗斯开发者非常珍视与你们的合作！🚀

GitHub Issue Template (English Version)

Title:
🚨 PaddleOCR Fails to Recognize Cyrillic (Russian) — Output Shows Transliteration

Description:
Hello! First, thank you for this amazing PaddleOCR tool. 🇨🇳🤝🇷🇺

I'm encountering an issue with Russian (Cyrillic) text recognition. The model runs but outputs transliterated Latin letters instead of proper Cyrillic characters.

How to Reproduce

Input Code:

from paddleocr import PaddleOCR
ocr = PaddleOCR(lang="ru")  
img_path = "https://book.sibverk.ru/image/trumb/500x500/13-921550.jpg"  
result = ocr.predict(img_path)  
for line in result:  
    print(line['rec_texts'])  

Actual Output:

['TMMH POCCNMCKOM 中EAEPALMN']

Expected Output:

['ГИМН РОССИЙСКОЙ ФЕДЕРАЦИИ']

Troubleshooting Attempted

1. Installed strictly per official docs:

   python -m pip install paddlepaddle-gpu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
   pip install paddleocr==3.0.0

2. Verified lang="ru" parameter is set
3. Tested with multiple images — issue persists

Questions

1. Is this a known issue? Are there solutions?
2. Are additional language models required for Cyrillic?
3. Are there post-processing parameters to correct transliteration?

Environment

• OS: (e.g., Ubuntu 22.04)
• Python: (e.g., 3.10.12)
• PaddlePaddle: 3.0.0
• PaddleOCR: 3.0.0

Thank you sincerely for your help! Chinese developers' work is truly admirable, and we Russian developers deeply value this collaboration. 🚀

[AdosBorodoS]

Hi! 👋 I just ran into exactly the same issue when trying to use lang="ru" with PaddleOCR — instead of proper Cyrillic letters, I’m getting Latin transliterations in the output (e.g., “TMMH POCCNMCKOM” instead of “ГИМН РОССИЙСКОЙ”).

I’ve followed all the setup instructions carefully and also tested with several images, but the problem persists.

Could a developer please take a look or advise if there’s any known fix for this? 🙏
Really hoping to use PaddleOCR for a Russian-language project and would be grateful for your help!

Thanks in advance!

[DVDinTheWoods]

Also duplicate it here just in case.I've started using this tool in my small project — so far everything is going very well. However, I’ve come across a similar issue with Cyrillic characters. Is there any solution available? Maybe there are any special settings for proper operation with the Russian language?

Would appreciate any help

[Godnoken]

I think the model simply just isn't good enough. All the other non-CJK & non-English models are equally broken.

[julesarts]

It may not be that complex. In my investigation, I found that using a Chinese model to detect other languages, such as Bangla, even without a publicly available detection model for that language, can yield good results. For your question, you simply need to create a custom dataset and fine-tune the model. 🤔🤔🤔

[zhangyubo0722]

Hello, Russian is already supported. Please refer to the documentation at https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.en.md and use the model eslav_PP-OCRv5_mobile_rec.