PaddleOCR/main/en/index

[giscus[bot]]

PaddleOCR/main/en/index

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)

https://paddlepaddle.github.io/PaddleOCR/main/en/index.html

[deftydev]

When paddle ocr will be supported by new python version , is there any plan to support newer python versions?

[GreatV]

Which Python version are you referring to?

[Liyulingyue]

It my depend on the dependency of ppocr.

[PatelKartik-lgm]

Hi
Could you please connect with your Developer or any demo person, I need brief details about this.

[Saranga99]

Anyone working with tabular data? wanted to extract different tables in a pdf file witch has around 40 pages and 50+ tables

[liuhongen1234567]

Hello, if you want to extract tables from a PDF, you can try the Table Recognition V2 pipeline at https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/pipeline_usage/table_recognition_v2.html.

[jmoore34]

Is there a library for using PaddleOCR 3.0 on Android? For example, with Google ML Kit you just add the library:

dependencies {
  implementation 'com.google.mlkit:text-recognition-chinese:16.0.1'
}

and then call it like so

val recognizer = TextRecognition.getClient(ChineseTextRecognizerOptions.Builder().build())
val image = InputImage.fromBitmap(bitmap, 0)
recognizer.process(image)
  .addOnSuccessListener { visionText ->
      // Task completed successfully
  }
  .addOnFailureListener { e ->
      // Task failed with an exception
  }

Would something similar be possible with PaddleOCR 3.0?

[BichengFan]

PP-OCRv5 检测/识别模型可以下载？

[liuhongen1234567]

您好，PP-OCRv5 检测/识别模型可以到PP-OCRv5文档的模型列表中下载，如果您想要在自己数据集上微调可以下载训练模型，如果希望直接使用paddleocr wheel包推理，可以下载推理模型

[BichengFan]

使用paddlex镜像，把模型转成onnx：
python -m paddlex --paddle2onnx --paddle_model_dir ./PP-OCRv5_server_det_infer --onnx_model_dir ./PP-OCR-DET-ONNX --opset_version 7
Input dir: PP-OCRv5_server_det_infer
Output dir: PP-OCR-DET-ONNX
Paddle2ONNX conversion starting...
grep: warning: GREP_OPTIONS is deprecated; please use an alias or script
[Paddle2ONNX] Start parsing the Paddle model file...
[Paddle2ONNX] Use opset_version = 11 for ONNX export.
[Paddle2ONNX] PaddlePaddle model is exported as ONNX format now.
2025-07-18 10:04:24 [INFO] Try to perform constant folding on the ONNX model with Polygraphy.
[W] 'colored' module is not installed, will not use colors when logging. To enable colors, please install the 'colored' module: python3 -m pip install colored
[I] Module: 'onnxruntime' is required, but not installed. Attempting to install now.
[I] Running installation command: /usr/bin/python -m pip install onnxruntime>=1.10.0
[I] Folding Constants | Pass 1
[I] Module: 'onnx_graphsurgeon' is required, but not installed. Attempting to install now.
[I] Running installation command: /usr/bin/python -m pip install onnx_graphsurgeon>=0.3.27 --extra-index-url=https://pypi.ngc.nvidia.com
[I] Total Nodes | Original: 1306, After Folding: 596 | 710 Nodes Folded
[I] Folding Constants | Pass 2
[I] Total Nodes | Original: 596, After Folding: 596 | 0 Nodes Folded
2025-07-18 10:05:13 [INFO] ONNX model saved in PP-OCR-DET-ONNX/inference.onnx.
Paddle2ONNX conversion succeeded
Copied PP-OCRv5_server_det_infer/inference.yml to PP-OCR-DET-ONNX/inference.yml
Done

使用onnxruntime(1.10.0)加载模型，
import numpy as np
import torch
import onnxruntime as ort

sess_options = ort.SessionOptions()
providers = ['CPUExecutionProvider']
detection_onnx_path = 'paddleOCR/detection/inference.onnx'
session = ort.InferenceSession(detection_onnx_path, sess_options=sess_options, providers=providers)
[ONNXRuntimeError] : 7 : INVALID_PROTOBUF : Load model from paddleOCR/detection/inference.onnx failed:Protobuf parsing failed.

请问是不是转onnx模型的时候，没成功？

[PiggyOverFlow]

请问PP-OCRV5的DB检测头，是否支持CTW-1500 或者Total-Test 数据集的训练

[liuhongen1234567]

您好，可以的，DBNet本身就是基于分割的方法，所以即使文字不规则也不影响训练。

[PiggyOverFlow]

感谢您的回复，我使用的是PP-OCRV5在下的configs/det/PP-OCRv5/PP-OCRv5_server_det.yml配置文件，当使用CTW1500或者Total-test数据集训练模型时，训练会出现卡死，详细的情况已在 issue( “https://github.com/PaddlePaddle/PaddleOCR/issues/15894”)
中描述。
期待您的回复

[AnugyaSahu]

Hi, I have a bunch of invoices in German language! and wanted to extract specific key information from them. every invoice has a different layout structure but I want the model to understand what it is searching for inside those complex structures, any ideas of what kind of pipeline should I build

[liuhongen1234567]

Hello, if you want to perform layout structure analysis, you can use the PP-Structurev3 model and then input the results of the layout analysis into large models such as GPT-4o. Alternatively, you can try using PP-ChatOCRv4, which is specifically designed for extraction tasks and has built-in large model integration. For the German language, simply replace the OCR model in the above pipeline with latin_PP-OCRv5_mobile_rec.

The relevant documentation is as follows:
PP-Structurev3 ： https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html
PP-ChatOCRv4： https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-ChatOCRv4.html
latin_PP-OCRv5_mobile_rec： https://www.paddleocr.ai/latest/en/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html

[ctgarvey]

Hi , is there perhaps a plain vanilla paddleocr 5 script and test image available for testing on my python rig? I cannot get this library to extract any text at all, even though tesseract and easyocr can extract plenty of text. I have tried using super simple ocr = PaddleOCR(use_angle_cls=True, lang='en') , result = ocr.predictor(proc)..... but it doesnt extract anything. ChatGPT gets completely confused with the paddleocr options, mixing up all the versions, and cannot generate anything that works. Is thre a limit on image size? i was trying with image (3512, 4968, 3). Many thanks!

[liuhongen1234567]

Hello, since (3512, 4968, 3) is a relatively large size and exceeds the default training configuration, you can resize the image to within 960, for example:

ocr = PaddleOCR(use_angle_cls=True, lang='en',text_det_limit_type='max',text_det_limit_side_len=960)

For more parameters, you can refer to the PP-OCRv5 documentation.
https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html