Skip to main content

AgentOCR Package 使用说明

1 快速上手

1.1 安装

  • pip 安装:

    # 安装 AgentOCR
    $ pip install agentocr 
    
    # 根据设备平台安装合适版本的 ONNXRuntime
    $ pip install onnxruntime
    
  • whl 包安装:

    • 下载 whl 包:链接

    • 安装 whl 包:

      # 安装 AgentOCR
      $ pip install agentocr-x.x.x-py3-none-any.whl 
      
      # 根据设备平台安装合适版本的 ONNXRuntime
      $ pip install onnxruntime
      
  • 源码安装:

    # 同步 AgentOCR 代码
    $ git clone https://github.com/AgentMaker/AgentOCR
    
    # 安装 AgentOCR 
    $ cd AgentOCR && python setup.py install
    
    # 根据设备平台安装合适版本的 ONNXRuntime
    $ pip install onnxruntime
    

2 使用

AgentOCR Package 会自动下载 PaddleOCR 中/英文轻量级模型作为默认模型
可通过切换其他内置配置文件或自定义配置文件进行模型和参数自定义
在 API 接口的使用上,基本和 PPOCR Package 保持一致

2.1 代码使用

2.1.1 API 接口

  • 接口介绍:

    class OCRSystem:
        
        def __init__(self, config='ch', warmup=True, **kwargs):
            '''
            The Inference OCR System of AgentOCR.
    
            Params:
                config: 配置文件名称或路径, 默认为 'ch'.
                warmup: 初始化时进行模型预热, 默认为 True.
                **kwargs: 更多配置参数,这些参数会覆盖配置文件中的相同配置.
            '''
    
        def ocr(self, img, det=True, cls=False, rec=True, return_cls=False):
            '''
            Params:
                img: 图片路径或者图片数组.
                det: 文本位置定位, 默认为 True.
                cls: 文本方向分类, 默认为 False.
                rec: 文本内容识别, 默认为 True.
                return_cls: 返回文本方向分类结果, 默认为 False.
            '''
    

2.1.2 使用样例

  • 检测 + 方向分类器 + 识别全流程

    from agentocr import OCRSystem
    
    # 通过 config 参数来进行模型配置,内置多国语言的配置文件
    # 也可以根据 3.1 配置文件中的选项,指定自定义配置,比如通过下面的代码指定运行后端:
    # ocr = OCRSystem(config='ch', providers='cuda,cpu')
    ocr = OCRSystem(config='ch')
    
    # 设置测试图片路径
    img_path = 'test.jpg'
    
    # 调用 OCR API 进行全流程识别
    result = ocr.ocr(img_path, cls=True)
    
      # 结果是一个list,每个item包含了文本框,文字和识别置信度
      [[[24.0, 36.0], [304.0, 34.0], [304.0, 72.0], [24.0, 74.0]], ['纯臻营养护发素', 0.964739]]
      [[[24.0, 80.0], [172.0, 80.0], [172.0, 104.0], [24.0, 104.0]], ['产品信息/参数', 0.98069626]]
      [[[24.0, 109.0], [333.0, 109.0], [333.0, 136.0], [24.0, 136.0]], ['(45元/每公斤,100公斤起订)', 0.9676722]]
      ......
    
  • 检测 + 识别

    from agentocr import OCRSystem
    
    ocr = OCRSystem(config='ch')
    
    img_path = 'test.jpg'
    
    # 关闭分类选项
    result = ocr.ocr(img_path, cls=False)
    
      # 结果是一个list,每个item包含了文本框,文字和识别置信度
      [[[24.0, 36.0], [304.0, 34.0], [304.0, 72.0], [24.0, 74.0]], ['纯臻营养护发素', 0.964739]]
      [[[24.0, 80.0], [172.0, 80.0], [172.0, 104.0], [24.0, 104.0]], ['产品信息/参数', 0.98069626]]
      [[[24.0, 109.0], [333.0, 109.0], [333.0, 136.0], [24.0, 136.0]], ['(45元/每公斤,100公斤起订)', 0.9676722]]
      ......
    
  • 方向分类器 + 识别

    from agentocr import OCRSystem
    
    ocr = OCRSystem(config='ch')
    
    img_path = 'test.jpg'
    
    # 关闭检测并开启分类选项
    result = ocr.ocr(img_path, det=False, cls=True)
    
      # 结果是一个list,每个item只包含识别结果和识别置信度
    
      ['韩国小馆', 0.9907421]
    
  • 单独执行检测

    from agentocr import OCRSystem
    
    ocr = OCRSystem(config='ch')
    
    img_path = 'test.jpg'
    
    # 关闭识别选项
    result = ocr.ocr(img_path, rec=False)
    
      # 结果是一个list,每个item只包含文本框
      [[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
      [[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
      [[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
      ......
    
  • 单独执行识别

    from agentocr import OCRSystem
    
    ocr = OCRSystem(config='ch')
    
    img_path = 'test.jpg'
    
    # 关闭检测选项
    result = ocr.ocr(img_path, det=False)
    
      # 结果是一个list,每个item只包含识别结果和识别置信度
      ['韩国小馆', 0.9907421]
    
  • 单独执行方向分类器

    from agentocr import OCRSystem
    
    ocr = OCRSystem(config='ch')
    
    img_path = 'test.jpg'
    
    # 关闭检测、识别并开启分类选项
    result = ocr.ocr(img_path, det=False, cls=True, rec=False)
    
      # 结果是一个list,每个item只包含分类结果和分类置信度
      ['0', 0.9999924]
    

2.2 服务器部署

2.2.1 启动 OCR 服务

  • 通过命令行启动:

    # config 配置文件 / host 监听地址 / port 监听接口 / 其他配置参数
    $ agentocr server \
        --config ch \
        --host 127.0.0.1 \
        --port 5000 \
        --providers cpu
    

2.2.2 接口调用

  • 接口地址:http://{host}:{port}/ocr

  • 请求类型:Post

  • 使用 Python 调用 OCR 服务接口:

    import cv2
    import json
    import base64
    import requests
    
    # 图片 Base64 编码
    def cv2_to_base64(image):
        data = cv2.imencode('.jpg', image)[1]
        image_base64 = base64.b64encode(data.tobytes()).decode('UTF-8')
        return image_base64
    
    
    # 读取图片
    image = cv2.imread('test.jpg')
    image_base64 = cv2_to_base64(image)
    
    # 构建请求数据
    data = {
        'image': image_base64,
        'det': True,
        'cls': True,
        'rec': True
    }
    
    # 发送请求
    url = "http://127.0.0.1:5000/ocr"
    r = requests.post(url=url, data=json.dumps(data))
    
    # 打印预测结果
    print(r.json())
    

3 配置

AgentOCR 使用 json 格式的配置文件来配置各种模型和各项参数
也同时内置了多个语言的预设参数配置,可通过对应的语言缩写进行快速调用
具体的语言和缩写的对应情况,请参考主页的 【预设配置选项】 表格
更多 PPOCR 的预训练模型,请跳转至主页 【预训练模型】 处下载

3.1 配置文件

  • 快速配置:

    可通过如下几个选项快速配置不同的模型文件、字典和可视化字体

    {
        "det_model_dir": "ch_ppocr_mobile_v2.0_det",
        "rec_model_dir": "ch_ppocr_mobile_v2.0_rec",
        "rec_char_type": "ch",
        "rec_char_dict_path": "ppocr_keys_v1",
        "vis_font_path": "simfang",
        "cls_model_dir": "ch_ppocr_mobile_v2.0_cls"
    }
    
  • 完整配置:

    详细的参数介绍请参考下一小节的内容

    {
        "providers": "auto",
        "det_algorithm": "DB",
        "det_model_dir": "ch_ppocr_mobile_v2.0_det",
        "det_limit_side_len": 960,
        "det_limit_type": "max",
        "det_db_thresh": 0.3,
        "det_db_box_thresh": 0.6,
        "det_db_unclip_ratio": 1.5,
        "use_dilation": false,
        "det_db_score_mode": "fast",
        "det_east_score_thresh": 0.8,
        "det_east_cover_thresh": 0.1,
        "det_east_nms_thresh": 0.2,
        "det_sast_score_thresh": 0.5,
        "det_sast_nms_thresh": 0.2,
        "det_sast_polygon": false,
        "rec_algorithm": "CRNN",
        "rec_model_dir": "ch_ppocr_mobile_v2.0_rec",
        "rec_image_shape": "3, 32, 320",
        "rec_char_type": "ch",
        "rec_batch_num": 8,
        "max_text_length": 25,
        "rec_char_dict_path": "ppocr_keys_v1",
        "use_space_char": true,
        "vis_font_path": "simfang",
        "drop_score": 0.5,
        "cls_model_dir": "ch_ppocr_mobile_v2.0_cls",
        "cls_image_shape": "3, 48, 192",
        "label_list": ["0", "180"],
        "cls_batch_num": 8,
        "cls_thresh": 0.9,
        "total_process_num": 1,
        "show_log": true
    }
    

3.2 参数说明

字段 说明 默认值
providers 使用的计算后端,使用字符串设定使用顺序,使用逗号分隔,不区分大小写,如 “cuda,cpu”,默认自动选择所有可用后端(可能并非最佳顺序) auto
det_algorithm 使用的检测算法类型 DB
det_model_dir 检测模型文件,文件名或文件路径 ch_ppocr_mobile_v2.0_det
det_max_side_len 检测算法前向时图片长边的最大尺寸,当长边超出这个值时会将长边resize到这个大小,短边等比例缩放 960
det_limit_type 检测的限制类型 max
det_db_thresh DB模型输出预测图的二值化阈值 0.3
det_db_box_thresh DB模型输出框的阈值,低于此值的预测框会被丢弃 0.6
det_db_unclip_ratio DB模型输出框扩大的比例 1.5
use_dilation 是否使用空洞卷积 False
det_db_score_mode DB 分数计算模式(slow or fast) fast
det_east_score_thresh EAST模型输出预测图的二值化阈值 0.8
det_east_cover_thresh EAST模型输出框的阈值,低于此值的预测框会被丢弃 0.1
det_east_nms_thresh EAST模型输出框NMS的阈值 0.2
det_sast_polygon 是否使用 SAST polygon False
rec_algorithm 使用的识别算法类型 CRNN
rec_model_dir 识别模型文件,文件名或文件路径 ch_ppocr_mobile_v2.0_rec
rec_image_shape 识别算法的输入图片尺寸 3,32,320
rec_char_type 识别算法的字符类型,中英文(ch)、英文(en)、法语(french)、德语(german)、韩语(korean)、日语(japan) ch
rec_batch_num 进行识别时,同时前向的图片数 8
max_text_length 识别算法能识别的最大文字长度 25
rec_char_dict_path 识别模型字典,文件名或文件路径 ppocr_keys_v1
use_space_char 是否识别空格 True
vis_font_path 可视化字体,文件名或者文件路径 simfang
drop_score 对输出按照分数(来自于识别模型)进行过滤,低于此分数的不返回 0.5
cls_model_dir 分类模型文件(文件名或者文件路径) ch_ppocr_mobile_v2.0_cls
cls_image_shape 分类算法的输入图片尺寸 3, 48, 192
label_list 分类算法的标签列表 ['0', '180']
cls_batch_num 进行分类时,同时前向的图片数 8
cls_thresh 分类器阈值 0.9
total_process_num 进程数量 1
show_log 是否打印 log False

Metadata

Release files for agentocr 2.0.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for agentocr 2.0.0
File Size Uploaded Provenance Actions
agentocr-2.0.0.tar.gz 95.9 kB Without provenance Details for agentocr-2.0.0.tar.gz

Built distributions (wheels)

Table of built distributions (wheels) for agentocr 2.0.0
File Interpreter ABI Platform Actions
agentocr-2.0.0-py3.9.egg Legacy Egg format - - Details for agentocr-2.0.0-py3.9.egg
agentocr-2.0.0-py3-none-any.whl Python 3 none any Details for agentocr-2.0.0-py3-none-any.whl

Total release size: 361.8 kB

Release files / agentocr-2.0.0.tar.gz

Download URL agentocr-2.0.0.tar.gz
Size 95.9 kB
Tags Source
SHA-256 checksum
How to use checksums
a24f5af81b06f87e05b528a674034bac74a81c29693a71e7621617f0f84fe2fd
BLAKE2b-256 checksum
How to use checksums
5d8bffab1e06925d69c88bfb74eebcc0d028902cd35c180acd3666b3863b5078
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/3.4.2 importlib_metadata/4.8.1 pkginfo/1.7.1 requests/2.26.0 requests-toolbelt/0.9.1 tqdm/4.62.3 CPython/3.9.7

Release files / agentocr-2.0.0-py3.9.egg

Download URL agentocr-2.0.0-py3.9.egg
Size 158.3 kB
Tags Egg
SHA-256 checksum
How to use checksums
8567436d1501dfb1faa6ac1403cb2f77a5a0cff3a2f31af67303c0ed546cb7dc
BLAKE2b-256 checksum
How to use checksums
7543ae7535674cf6bbafb104ba83f3bf493d4e192147f9f79058db10302de6c6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/3.4.2 importlib_metadata/4.8.1 pkginfo/1.7.1 requests/2.26.0 requests-toolbelt/0.9.1 tqdm/4.62.3 CPython/3.9.7

Release files / agentocr-2.0.0-py3-none-any.whl

Download URL agentocr-2.0.0-py3-none-any.whl
Size 107.6 kB
Tags Python 3
SHA-256 checksum
How to use checksums
668f7124eeed10fcb421adbae0f148e6d22f226fae214e79d9579d5aff790e7b
BLAKE2b-256 checksum
How to use checksums
f002aef691c9c6994482c921ccc157a32d9c30b8643889a9babac38a0bf3e606
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/3.4.2 importlib_metadata/4.8.1 pkginfo/1.7.1 requests/2.26.0 requests-toolbelt/0.9.1 tqdm/4.62.3 CPython/3.9.7

Security for agentocr 2.0.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page