PaddleOCR2.7+Qt5

章节一:Windows 下的 PIP 安装

        官网安装教程地址

        按照里面的教程去安装

         如果使用cuda版本的还要安装tensorrt,不然后面运行demo程序的程序会报如下错。

        下载TensorRT 8版本,tensorrt下载地址

章节二:编译源码

        进入官网源码地址

        下载release2.7

        下载三个模型

        下载推理预测库

          官网下载地址,根据自己的情况下载CPU或对应的自己CUDA版本的GPU

用CMAK打开如下文件

    添加路径OpenCV路径和预测库路径。

图片

    然后Configers,再Generate,如果有报错不用管他,最后在你的构建目录生成了项目,然后开始编译。

        但是编译会报错

在utility.cpp中无法打开包括文件"dirent.h":No such file or directory

        <dirent.h>是个unix系统下常见的接口,但windows平台的MSVC编译器并没有提供这个接口,对于跨平台的项目开发就会带来一些麻烦,如果在MSVC下编译时可能因为windows平台缺少这个接口就要为windows平台另外写一些代码。

        不过大佬已经做了一个windows版本的<dirent.h>,放在了github上面,链接如下:

https://github.com/tronkko/dirent

        下载完后加入这个文件夹:

        然后重新编译,但是会报错找不到_stat,做如下修改就可以了。

//修改前
    struct stat s;
    _stat(dir_name, &s);
//修改后
    struct _stat64 s;
    _stat64(dir_name, &s);

        最后就可以生成成功,生成成功要将这几个dll复制到程序目录。

        onnxruntime和paddle2onnx在这里

        将这个文件复制到程序目录

        源码里面这个文件的路径也要改一下

        然后输入命令运行程序,下面的路径根据自己情况去修改

./ppocr.exe --det_model_dir=F:/Vision/PaddleOCR/2_7/model/ch_PP-OCRv4_det_infer --cls_model_dir=F:/Vision/PaddleOCR/2_7/model/ch_ppocr_mobile_v2.0_cls_infer --rec_model_dir=F:/Vision/PaddleOCR/2_7/model/ch_PP-OCRv4_rec_infer --image_dir=F:/Vision/PaddleOCR/2_7/PaddleOCR-release-2.7/deploy/cpp_infer/build/Release/aa.png

        运行结果如下

章节三:在Qt上编译运行

        pro文件构建

QT       += core gui widgets

greaterThan(QT_MAJOR_VERSION, 4): QT += widgets

CONFIG += c++17

DEFINES += QT_DEPRECATED_WARNINGS

##################################################################
#指定生成的文件存放位置
##################################################################
MOC_DIR         = $$PWD/temp/moc
RCC_DIR         = $$PWD/temp/rcc
UI_DIR          = $$PWD/temp/ui
OBJECTS_DIR     = $$PWD/temp/obj
DESTDIR         = $$PWD/temp/bin

CONFIG(debug, debug|release) {
    QMAKE_CXXFLAGS_DEBUG += /MTd
}

CONFIG(release, debug|release) {
    QMAKE_CXXFLAGS_RELEASE += /MT
}

PaddleOCR_ROOT = F:/Vision/PaddleOCR/2_7/OCR_Demo/OCR_Demo/ocr
SOURCES += \
#    $$PaddleOCR_ROOT/src/args.cpp \
    $$PaddleOCR_ROOT/src/clipper.cpp \
    $$PaddleOCR_ROOT/src/ocr_cls.cpp \
    $$PaddleOCR_ROOT/src/ocr_det.cpp \
    $$PaddleOCR_ROOT/src/ocr_rec.cpp \
#    $$PaddleOCR_ROOT/src/paddleocr.cpp \
#    $$PaddleOCR_ROOT/src/paddlestructure.cpp \
    $$PaddleOCR_ROOT/src/postprocess_op.cpp \
    $$PaddleOCR_ROOT/src/preprocess_op.cpp \
    $$PaddleOCR_ROOT/src/structure_layout.cpp \
    $$PaddleOCR_ROOT/src/structure_table.cpp \
    $$PaddleOCR_ROOT/src/utility.cpp \
    ScreenWidget/screen.cpp \
    ScreenWidget/screenwidget.cpp \
    main.cpp \
    mainwindow.cpp \
    my_config.cpp \
    my_paddleocr.cpp

HEADERS += \
    $$PaddleOCR_ROOT/include/ocr_cls.h \
    $$PaddleOCR_ROOT/include/ocr_det.h \
    $$PaddleOCR_ROOT/include/ocr_rec.h \
#    $$PaddleOCR_ROOT/include/paddleocr.h \
#    $$PaddleOCR_ROOT/include/paddlestructure.h \
    $$PaddleOCR_ROOT/include/postprocess_op.h \
    $$PaddleOCR_ROOT/include/preprocess_op.h \
    $$PaddleOCR_ROOT/include/structure_layout.h \
    $$PaddleOCR_ROOT/include/structure_table.h \
    $$PaddleOCR_ROOT/include/utility.h \
#    $$PaddleOCR_ROOT/include/args.h \
    $$PaddleOCR_ROOT/include/clipper.h \
    $$PaddleOCR_ROOT/include/dirent.h \
    ScreenWidget/screen.h \
    ScreenWidget/screenwidget.h \
    mainwindow.h \
    my_config.h \
    my_paddleocr.h

FORMS += \
    mainwindow.ui

INCLUDEPATH += $$PaddleOCR_ROOT
INCLUDEPATH += $$PaddleOCR_ROOT/include
INCLUDEPATH += $$PWD\ScreenWidget

Inference_ROOT = F:/Vision/PaddleOCR/2_7/prelib/cuda11_0
INCLUDEPATH += $$Inference_ROOT/paddle/include
INCLUDEPATH += $$Inference_ROOT/third_party/install/protobuf/include
INCLUDEPATH += $$Inference_ROOT/third_party/install/glog/include
#INCLUDEPATH += $$Inference_ROOT/third_party/install/gflags/include
INCLUDEPATH += $$Inference_ROOT/third_party/install/xxhash/include
INCLUDEPATH += $$Inference_ROOT/third_party/install/mklml/include
INCLUDEPATH += $$Inference_ROOT/third_party/install/mkldnn/include

LIBS += -L$$Inference_ROOT/paddle/lib -lpaddle_inference
LIBS += -L$$Inference_ROOT/third_party/install/mklml/lib -lmklml
LIBS += -L$$Inference_ROOT/third_party/install/mklml/lib -llibiomp5md
LIBS += -L$$Inference_ROOT/third_party/install/mkldnn/lib -lmkldnn
LIBS += -L$$Inference_ROOT/third_party/install/glog/lib -lglog
#LIBS += -L$$Inference_ROOT/third_party/install/gflags/lib -lgflags_static
LIBS += -L$$Inference_ROOT/third_party/install/protobuf/lib -llibprotobuf
LIBS += -L$$Inference_ROOT/third_party/install/xxhash/lib -lxxhash

OpenCV_ROOT = E:/2021_software/OpenCv/OpenCv3_4_15/install/opencv/build
INCLUDEPATH += $$OpenCV_ROOT/include
INCLUDEPATH += $$OpenCV_ROOT/include/opencv
INCLUDEPATH += $$OpenCV_ROOT/include/opencv2
LIBS += -L$$OpenCV_ROOT/x64/vc15/lib -lopencv_world3415

        自定义一个config类

#pragma once

#include <iomanip>
#include <iostream>
#include <map>
#include <ostream>
#include <string>
#include <vector>

#include "include/utility.h"

using namespace PaddleOCR;

class MY_OCRConfig {
public:
  explicit MY_OCRConfig(const std::string &config_file);

    // common args
    bool use_gpu = false;
    bool use_tensorrt = false;
    int gpu_id = 0;
    int gpu_mem = 4000;
    int cpu_threads = 10;
    bool enable_mkldnn = false;
    std::string precision = "fp32";
    bool benchmark = false;
    std::string output = "./output/";
    std::string image_dir = "";
    std::string type = "ocr";

    // detection related
    std::string det_model_dir = "";
    std::string limit_type = "max";
    int limit_side_len = 960;
    double det_db_thresh = 0.3;
    double det_db_box_thresh = 0.6;
    double det_db_unclip_ratio = 1.5;
    bool use_dilation = false;
    std::string det_db_score_mode = "slow";
    bool visualize = true;

    // classification related
    bool use_angle_cls = false;
    std::string cls_model_dir = "";
    double cls_thresh = 0.9;
    int cls_batch_num = 1;

    // recognition related
    std::string rec_model_dir = "";
    int rec_batch_num = 6;
    std::string rec_char_dict_path = "./ppocr_keys_v1.txt";
    int rec_img_h = 48;
    int rec_img_w = 320;

    // layout model related
    std::string layout_model_dir = "";
    std::string layout_dict_path = "../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt";
    double layout_score_threshold = 0.5;
    double layout_nms_threshold = 0.5;

    // structure model related
    std::string table_model_dir = "";
    int table_max_len = 488;
    int table_batch_num = 1;
    bool merge_no_span_structure = true;
    std::string table_char_dict_path = "../../ppocr/utils/dict/table_structure_dict_ch.txt";

    // ocr forward related
    bool det = true;
    bool rec = true;
    bool cls = false;
    bool table = false;
    bool layout = false;


private:
  // Load configuration
  std::map<std::string, std::string> LoadConfig(const std::string &config_file);

  std::vector<std::string> split(const std::string &str,
                                 const std::string &delim);

  std::map<std::string, std::string> config_map_;
};

#include "my_config.h"
#include <qdebug.h>

std::vector<std::string> MY_OCRConfig::split(const std::string &str,
                                          const std::string &delim) {
  std::vector<std::string> res;
  if ("" == str)
    return res;
  int strlen = str.length() + 1;
  char *strs = new char[strlen];
  std::strcpy(strs, str.c_str());

  int delimlen = delim.length() + 1;
  char *d = new char[delimlen];
  std::strcpy(d, delim.c_str());

  char *p = std::strtok(strs, d);
  while (p) {
    std::string s = p;
    res.push_back(s);
    p = std::strtok(NULL, d);
  }

  delete[] strs;
  delete[] d;
  return res;
}

std::map<std::string, std::string>
MY_OCRConfig::LoadConfig(const std::string &config_path) {
  auto config = Utility::ReadDict(config_path);

  std::map<std::string, std::string> dict;
  for (int i = 0; i < config.size(); i++) {
    // pass for empty line or comment
    if (config[i].size() <= 1 || config[i][0] == '#') {
      continue;
    }

    //
    std::vector<std::string> res = split(config[i], " ");
    if (res.size() < 2) {
        dict[res[0]] = "";
    }else{
        dict[res[0]] = res[1];
    }
  }
  return dict;
}

MY_OCRConfig::MY_OCRConfig(const std::string &config_file)
{
   config_map_ = LoadConfig(config_file);

   // common args
   this->use_gpu        = (config_map_["use_gpu"] == "true");
   this->use_tensorrt   = (config_map_["use_tensorrt"] == "true");
   this->gpu_id         = stoi(config_map_["gpu_id"]);
   this->gpu_mem        = stoi(config_map_["gpu_mem"]);
   this->cpu_threads    = stoi(config_map_["cpu_threads"]);
   this->enable_mkldnn  = (config_map_["enable_mkldnn"] == "true");
   this->precision      = config_map_["precision"];
   this->benchmark      = (config_map_["benchmark"] == "true");
   this->output         = config_map_["output"];
   this->image_dir      = config_map_["image_dir"];
   this->type           = config_map_["type"];

   // detection related
   this->det_model_dir       = config_map_["det_model_dir"];
   this->limit_type          = config_map_["limit_type"];
   this->limit_side_len      = stoi(config_map_["limit_side_len"]);
   this->det_db_thresh       = stod(config_map_["det_db_thresh"]);
   this->det_db_box_thresh   = stod(config_map_["det_db_box_thresh"]);
   this->det_db_unclip_ratio = stod(config_map_["det_db_unclip_ratio"]);
   this->use_dilation        = (config_map_["use_dilation"] == "true");
   this->det_db_score_mode   = config_map_["det_db_score_mode"];
   this->visualize           = (config_map_["visualize"] == "true");

   // classification related
   this->use_angle_cls   = (config_map_["use_angle_cls"] == "true");
   this->cls_model_dir   = config_map_["cls_model_dir"];
   this->cls_thresh      = stod(config_map_["cls_thresh"]);
   this->cls_batch_num   = stoi(config_map_["cls_batch_num"]);

   // recognition related
   this->rec_model_dir        = config_map_["rec_model_dir"];
   this->rec_batch_num        = stoi(config_map_["rec_batch_num"]);
   this->rec_char_dict_path   = config_map_["rec_char_dict_path"];
   this->rec_img_h            = stoi(config_map_["rec_img_h"]);
   this->rec_img_w            = stoi(config_map_["rec_img_w"]);

   // layout model related
   this->layout_model_dir        = config_map_["layout_model_dir"];
   this->layout_dict_path        = config_map_["layout_dict_path"];
   this->layout_score_threshold  = stod(config_map_["layout_score_threshold"]);
   this->layout_nms_threshold    = stod(config_map_["layout_nms_threshold"]);

   // structure model related
   this->table_model_dir           = config_map_["table_model_dir"];
   this->table_max_len             = stoi(config_map_["table_max_len"]);
   this->table_batch_num           = stoi(config_map_["table_batch_num"]);
   this->merge_no_span_structure   = (config_map_["merge_no_span_structure"] == "true");
   this->table_char_dict_path      = config_map_["table_char_dict_path"];

   // ocr forward related
   this->det     = (config_map_["det"] == "true");
   this->rec     = (config_map_["rec"] == "true");
   this->cls     = (config_map_["cls"] == "true");
   this->table   = (config_map_["table"] == "true");
   this->layout  = (config_map_["layout"] == "true");
   qDebug()<<this->det<<config_map_["det"].c_str()<<QString(config_map_["det"].c_str())<<this->rec<<this->cls<<this->rec_model_dir.c_str();
}

        自定义一个ocr类

#pragma once

#include <include/ocr_cls.h>
#include <include/ocr_det.h>
#include <include/ocr_rec.h>

#include "my_config.h"

using namespace PaddleOCR;
class MY_PPOCR {
public:
  explicit MY_PPOCR();
  ~MY_PPOCR();

  std::vector<std::vector<OCRPredictResult>> ocr(std::vector<cv::Mat> img_list,
                                                 bool det = true,
                                                 bool rec = true,
                                                 bool cls = true);
  std::vector<OCRPredictResult> ocr(cv::Mat img, bool det = true,
                                    bool rec = true, bool cls = true);

  void reset_timer();
  void benchmark_log(int img_num);
    MY_OCRConfig *p_config;
protected:
  std::vector<double> time_info_det = {0, 0, 0};
  std::vector<double> time_info_rec = {0, 0, 0};
  std::vector<double> time_info_cls = {0, 0, 0};

  void det(cv::Mat img, std::vector<OCRPredictResult> &ocr_results);
  void rec(std::vector<cv::Mat> img_list,
           std::vector<OCRPredictResult> &ocr_results);
  void cls(std::vector<cv::Mat> img_list,
           std::vector<OCRPredictResult> &ocr_results);

private:
  std::unique_ptr<DBDetector> detector_;
  std::unique_ptr<Classifier> classifier_;
  std::unique_ptr<CRNNRecognizer> recognizer_;
};

#include "my_paddleocr.h"
#include <qdebug.h>
//#include "auto_log/autolog.h"


MY_PPOCR::MY_PPOCR()
{
    qDebug()<<"aaa1";
    p_config = new MY_OCRConfig("./ocrconfig.txt");

    qDebug()<<"aaa2";
    //
    if (p_config->det) {
      this->detector_.reset(new DBDetector(
          p_config->det_model_dir, p_config->use_gpu, p_config->gpu_id, p_config->gpu_mem,
          p_config->cpu_threads, p_config->enable_mkldnn, p_config->limit_type,
          p_config->limit_side_len, p_config->det_db_thresh, p_config->det_db_box_thresh,
          p_config->det_db_unclip_ratio, p_config->det_db_score_mode, p_config->use_dilation,
          p_config->use_tensorrt, p_config->precision));
    }

    qDebug()<<"aaa3";
    //
    if (p_config->cls && p_config->use_angle_cls) {
      this->classifier_.reset(new Classifier(
          p_config->cls_model_dir, p_config->use_gpu, p_config->gpu_id, p_config->gpu_mem,
          p_config->cpu_threads, p_config->enable_mkldnn, p_config->cls_thresh,
          p_config->use_tensorrt, p_config->precision, p_config->cls_batch_num));
    }
    qDebug()<<"aaa4";

    //
    if (p_config->rec) {
      this->recognizer_.reset(new CRNNRecognizer(
          p_config->rec_model_dir, p_config->use_gpu, p_config->gpu_id, p_config->gpu_mem,
          p_config->cpu_threads, p_config->enable_mkldnn, p_config->rec_char_dict_path,
          p_config->use_tensorrt, p_config->precision, p_config->rec_batch_num,
          p_config->rec_img_h, p_config->rec_img_w));
    }
    qDebug()<<"aaa5";
}

MY_PPOCR::~MY_PPOCR()
{
    delete p_config;
}

std::vector<std::vector<OCRPredictResult>>
MY_PPOCR::ocr(std::vector<cv::Mat> img_list, bool det, bool rec, bool cls)
{
    std::vector<std::vector<OCRPredictResult>> ocr_results;
    if (!det) {
        std::vector<OCRPredictResult> ocr_result;
        ocr_result.resize(img_list.size());
        if (cls && this->classifier_) {
            this->cls(img_list, ocr_result);
            for (int i = 0; i < img_list.size(); i++) {
                if (ocr_result[i].cls_label % 2 == 1 &&
                    ocr_result[i].cls_score > this->classifier_->cls_thresh) {
                    cv::rotate(img_list[i], img_list[i], 1);
                }
            }
        }
        if (rec) {
            this->rec(img_list, ocr_result);
        }
        for (int i = 0; i < ocr_result.size(); ++i) {
            std::vector<OCRPredictResult> ocr_result_tmp;
            ocr_result_tmp.push_back(ocr_result[i]);
            ocr_results.push_back(ocr_result_tmp);
        }
    } else {
      for (int i = 0; i < img_list.size(); ++i) {
        std::vector<OCRPredictResult> ocr_result =
            this->ocr(img_list[i], true, rec, cls);
        ocr_results.push_back(ocr_result);
      }
    }
    return ocr_results;
}

std::vector<OCRPredictResult> MY_PPOCR::ocr(cv::Mat img, bool det, bool rec, bool cls)
{
    std::vector<OCRPredictResult> ocr_result;
    // det
    this->det(img, ocr_result);
    // crop image
    std::vector<cv::Mat> img_list;
    for (int j = 0; j < ocr_result.size(); j++) {
        cv::Mat crop_img;
        crop_img = Utility::GetRotateCropImage(img, ocr_result[j].box);
        img_list.push_back(crop_img);
    }
    // cls
    if (cls && this->classifier_) {
        this->cls(img_list, ocr_result);
        for (int i = 0; i < img_list.size(); i++) {
          if (ocr_result[i].cls_label % 2 == 1 &&
              ocr_result[i].cls_score > this->classifier_->cls_thresh) {
              cv::rotate(img_list[i], img_list[i], 1);
          }
        }
    }
    // rec
    if (rec) {
        this->rec(img_list, ocr_result);
    }
    return ocr_result;
}

void MY_PPOCR::det(cv::Mat img, std::vector<OCRPredictResult> &ocr_results)
{
    std::vector<std::vector<std::vector<int>>> boxes;
    std::vector<double> det_times;

    this->detector_->Run(img, boxes, det_times);

    for (int i = 0; i < boxes.size(); i++) {
        OCRPredictResult res;
        res.box = boxes[i];
        ocr_results.push_back(res);
    }
    // sort boex from top to bottom, from left to right
    Utility::sorted_boxes(ocr_results);
    this->time_info_det[0] += det_times[0];
    this->time_info_det[1] += det_times[1];
    this->time_info_det[2] += det_times[2];
}

void MY_PPOCR::rec(std::vector<cv::Mat> img_list, std::vector<OCRPredictResult> &ocr_results)
{
    std::vector<std::string> rec_texts(img_list.size(), "");
    std::vector<float> rec_text_scores(img_list.size(), 0);
    std::vector<double> rec_times;
    this->recognizer_->Run(img_list, rec_texts, rec_text_scores, rec_times);
    // output rec results
    for (int i = 0; i < rec_texts.size(); i++) {
        ocr_results[i].text = rec_texts[i];
        ocr_results[i].score = rec_text_scores[i];
    }
    this->time_info_rec[0] += rec_times[0];
    this->time_info_rec[1] += rec_times[1];
    this->time_info_rec[2] += rec_times[2];
}

void MY_PPOCR::cls(std::vector<cv::Mat> img_list, std::vector<OCRPredictResult> &ocr_results)
{
    std::vector<int> cls_labels(img_list.size(), 0);
    std::vector<float> cls_scores(img_list.size(), 0);
    std::vector<double> cls_times;
    this->classifier_->Run(img_list, cls_labels, cls_scores, cls_times);
    // output cls results
    for (int i = 0; i < cls_labels.size(); i++) {
        ocr_results[i].cls_label = cls_labels[i];
        ocr_results[i].cls_score = cls_scores[i];
    }
    this->time_info_cls[0] += cls_times[0];
    this->time_info_cls[1] += cls_times[1];
    this->time_info_cls[2] += cls_times[2];
}

void MY_PPOCR::reset_timer()
{
    this->time_info_det = {0, 0, 0};
    this->time_info_rec = {0, 0, 0};
    this->time_info_cls = {0, 0, 0};
}

        核心调用代码


    //
    std::vector<cv::Mat> img_list;
    cv::Mat srcimg = cv::imread(qstr2str(fileName).data(), cv::IMREAD_COLOR);
    img_list.push_back(srcimg);
    p_ocr->reset_timer();

    //
    QElapsedTimer  RunTimer;
    RunTimer.start();
    std::vector<std::vector<OCRPredictResult>> ocr_results = p_ocr->ocr(img_list, p_ocr->p_config->det, p_ocr->p_config->rec, p_ocr->p_config->cls);
    ui->textBrowser->append(QString("检测时间:%1ms ---------------------").arg(RunTimer.elapsed()));

    //
    for (int i = 0; i < img_list.size(); ++i) {
        std::vector<OCRPredictResult> &ocr_result = ocr_results[i];
        for (int i = 0; i < ocr_result.size(); i++) {
            QString oustr;
            oustr += QString::number(i) + " ";

            // det
//            std::vector<std::vector<int>> boxes = ocr_result[i].box;
//            if (boxes.size() > 0) {
//                oustr += "det boxes: [";
//                for (int n = 0; n < boxes.size(); n++) {
//                    oustr +=  "[" + QString::number(boxes[n][0]) + "," + QString::number(boxes[n][1]) + "]";
//                    if (n != boxes.size() - 1) {
//                        oustr +=  ",";
//                    }
//                }
//                oustr +=  "]";
//            }

            // rec
            if (ocr_result[i].score != -1.0) {
                  oustr += " score: " + QString::number(ocr_result[i].score,'f',2) + " text: " + QString::fromUtf8(ocr_result[i].text.c_str()) + " ";
            }

            // cls
            if (ocr_result[i].cls_label != -1) {
              oustr += "cls label: " + QString::number(ocr_result[i].cls_label) + " cls score: " + ocr_result[i].cls_score;
            }

            //
//            oustr += "\r\n";
            ui->textBrowser->append(oustr);
        }

        //
        for (int n = 0; n < ocr_result.size(); n++) {
            cv::Point rook_points[4];

            //
            for (int m = 0; m < ocr_result[n].box.size(); m++) {
                rook_points[m] = cv::Point(int(ocr_result[n].box[m][0]), int(ocr_result[n].box[m][1]));
            }

            //
            const cv::Point *ppt[1] = {rook_points};
            int npt[] = {4};
            cv::polylines(img_list[i], ppt, npt, 1, 1, CV_RGB(255, 0, 0), 2, 8, 0);
        }

        //
        QImage outimage;
        cvMat2QImage(img_list[i], outimage);
        ui->label_image->setPixmap(QPixmap::fromImage(outimage.scaled(ui->label_image->width(),
                                                                      ui->label_image->height(),
                                                                      Qt::KeepAspectRatio)));
    }

        程序目录包含这个几个文件,如果你没有,在源码路径和预测库路径去搜索

        ocrconfig.txt是自定义的配置

# common args
use_gpu false
use_tensorrt false
gpu_id 0
gpu_mem 4000
cpu_threads 10
enable_mkldnn false
precision fp32
benchmark false
output ./output/
image_dir 
type ocr

# detection related
det_model_dir F:\Vision\PaddleOCR\2_7\model\ch_PP-OCRv4_det_infer
limit_type max
limit_side_len 960
det_db_thresh 0.3
det_db_box_thresh 0.6
det_db_unclip_ratio 1.5
use_dilation false
det_db_score_mode slow
visualize true

# classification related
use_angle_cls false
cls_model_dir F:\Vision\PaddleOCR\2_7\model\ch_ppocr_mobile_v2.0_cls_infer
cls_thresh 0.9
cls_batch_num 1

# recognition related
rec_model_dir F:\Vision\PaddleOCR\2_7\model\ch_PP-OCRv4_rec_infer
rec_batch_num 6
rec_char_dict_path ./ppocr_keys_v1.txt
rec_img_h 48
rec_img_w 320

# layout model related
layout_model_dir 
layout_dict_path ./layout_publaynet_dict.txt
layout_score_threshold 0.5
layout_nms_threshold 0.5

# structure model related
table_model_dir 
table_max_len 488
table_batch_num 1
merge_no_span_structure true
table_char_dict_path ./table_structure_dict_ch.txt

# ocr forward related
det true
rec true
cls false
table false
layout false

        最后运行结果

       有需要源码测试的, 这是源码地址

  • 14
    点赞
  • 14
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值