[Regression] Missing Spaces Between Words

[karanataryn]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

PaddleOCR 3.x struggles to identify spaces between words correctly, this is unlike behavior on similar models in 2.x. Similar issue to this earlier: #5448

🏃‍♂️ Environment (运行环境)

Linux, Debian 12 x86, Paddle 3.0.2

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

Please use the image below and run .predict() on it.

[liuhongen1234567]

Hello, you can try replacing the recognition model with a specialized English model, such as: en_PP-OCRv3_mobile_rec. The sample code and prediction results are as follows:

from paddleocr import PaddleOCR

ocr = PaddleOCR(
    text_recognition_model_name="en_PP-OCRv3_mobile_rec",
)
result = ocr.predict("/paddle/issue_test/demo.jpg")
for res in result:
    res.print()
    res.save_to_img("output")
    res.save_to_json("output")

[karanataryn]

Hello, you can try replacing the recognition model with a specialized English model, such as: en_PP-OCRv3_mobile_rec. The sample code and prediction results are as follows:

from paddleocr import PaddleOCR

ocr = PaddleOCR(
text_recognition_model_name="en_PP-OCRv3_mobile_rec",
)
result = ocr.predict("/paddle/issue_test/demo.jpg")
for res in result:
res.print()
res.save_to_img("output")
res.save_to_json("output")

Thanks, is that the model you recommend for English based use cases?

[liuhongen1234567]

Hello, you can test more cases to select the best model. I recommend using the latest PP-OCRv5 Latin language model in paddleocr 3.1 . Here is a sample code:

from paddleocr import PaddleOCR

ocr = PaddleOCR(
    text_recognition_model_name="latin_PP-OCRv5_mobile_rec", # 通过 lang 参数指定使用法语的识别模型
    use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 参数指定不使用文档方向分类模型
    use_doc_unwarping=False, # 通过 use_doc_unwarping 参数指定不使用文本图像矫正模型
    use_textline_orientation=False, # 通过 use_textline_orientation 参数指定不使用文本行方向分类模型
)
result = ocr.predict("/paddle/env/issue_test/1.jpg")
for res in result:
    res.print()
    res.save_to_img("output")
    res.save_to_json("output")

To use this model, you need to upgrade PaddleOCR to version 3.1.0.

 python -m pip install paddleocr==3.1.0

The documentation is as follows:
https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html#2-quick-start

[karanataryn]

Hello, you can test more cases to select the best model. I recommend using the latest PP-OCRv5 Latin language model in paddleocr 3.1 . Here is a sample code:

from paddleocr import PaddleOCR

ocr = PaddleOCR(
text_recognition_model_name="latin_PP-OCRv5_mobile_rec", # 通过 lang 参数指定使用法语的识别模型
use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 参数指定不使用文档方向分类模型
use_doc_unwarping=False, # 通过 use_doc_unwarping 参数指定不使用文本图像矫正模型
use_textline_orientation=False, # 通过 use_textline_orientation 参数指定不使用文本行方向分类模型
)
result = ocr.predict("/paddle/env/issue_test/1.jpg")
for res in result:
res.print()
res.save_to_img("output")
res.save_to_json("output")

To use this model, you need to upgrade PaddleOCR to version 3.1.0.

 python -m pip install paddleocr==3.1.0

The documentation is as follows: https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html#2-quick-start

This is helpful and fixes the issue, thanks! Closing this for now.