Skip to content

GitLab

"git@developer.sourcefind.cn:chenpangpang/ComfyUI.git" did not exist on "91ed2815d542c96fdad75edba2205140de3cbba6"
Commit aa59fca5 authored by Leif's avatar Leif
Browse files

Merge remote-tracking branch 'origin/dygraph' into dygraph

parents 12d15752 f01f24c7
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 1250 additions and 487 deletions
deploy/cpp_infer/src/args.cpp
View file @ aa59fca5
...@@ -47,6 +47,8 @@ DEFINE_string(rec_model_dir, "", "Path of rec inference model."); ...@@ -47,6 +47,8 @@ DEFINE_string(rec_model_dir, "", "Path of rec inference model.");
DEFINE_int32(rec_batch_num, 6, "rec_batch_num."); DEFINE_int32(rec_batch_num, 6, "rec_batch_num.");
DEFINE_string(rec_char_dict_path, "../../ppocr/utils/ppocr_keys_v1.txt", DEFINE_string(rec_char_dict_path, "../../ppocr/utils/ppocr_keys_v1.txt",
"Path of dictionary."); "Path of dictionary.");
DEFINE_int32(rec_img_h, 32, "rec image height");
DEFINE_int32(rec_img_w, 320, "rec image width");
// ocr forward related // ocr forward related
DEFINE_bool(det, true, "Whether use det in forward."); DEFINE_bool(det, true, "Whether use det in forward.");
......
deploy/cpp_infer/src/main.cpp
View file @ aa59fca5
...@@ -69,7 +69,7 @@ int main(int argc, char **argv) { ...@@ -69,7 +69,7 @@ int main(int argc, char **argv) {
cv::glob(FLAGS_image_dir, cv_all_img_names); cv::glob(FLAGS_image_dir, cv_all_img_names);
std::cout << "total images num: " << cv_all_img_names.size() << endl; std::cout << "total images num: " << cv_all_img_names.size() << endl;
PaddleOCR::PaddleOCR ocr = PaddleOCR::PaddleOCR(); PPOCR ocr = PPOCR();
std::vector<std::vector<OCRPredictResult>> ocr_results = std::vector<std::vector<OCRPredictResult>> ocr_results =
ocr.ocr(cv_all_img_names, FLAGS_det, FLAGS_rec, FLAGS_cls); ocr.ocr(cv_all_img_names, FLAGS_det, FLAGS_rec, FLAGS_cls);
......
deploy/cpp_infer/src/ocr_rec.cpp
View file @ aa59fca5
...@@ -39,7 +39,9 @@ void CRNNRecognizer::Run(std::vector<cv::Mat> img_list, ...@@ -39,7 +39,9 @@ void CRNNRecognizer::Run(std::vector<cv::Mat> img_list,
auto preprocess_start = std::chrono::steady_clock::now(); auto preprocess_start = std::chrono::steady_clock::now();
int end_img_no = min(img_num, beg_img_no + this->rec_batch_num_); int end_img_no = min(img_num, beg_img_no + this->rec_batch_num_);
int batch_num = end_img_no - beg_img_no; int batch_num = end_img_no - beg_img_no;
float max_wh_ratio = 0; int imgH = this->rec_image_shape_[1];
int imgW = this->rec_image_shape_[2];
float max_wh_ratio = imgW * 1.0 / imgH;
for (int ino = beg_img_no; ino < end_img_no; ino++) { for (int ino = beg_img_no; ino < end_img_no; ino++) {
int h = img_list[indices[ino]].rows; int h = img_list[indices[ino]].rows;
int w = img_list[indices[ino]].cols; int w = img_list[indices[ino]].cols;
...@@ -47,28 +49,28 @@ void CRNNRecognizer::Run(std::vector<cv::Mat> img_list, ...@@ -47,28 +49,28 @@ void CRNNRecognizer::Run(std::vector<cv::Mat> img_list,
max_wh_ratio = max(max_wh_ratio, wh_ratio); max_wh_ratio = max(max_wh_ratio, wh_ratio);
} }
int batch_width = 0; int batch_width = imgW;
std::vector<cv::Mat> norm_img_batch; std::vector<cv::Mat> norm_img_batch;
for (int ino = beg_img_no; ino < end_img_no; ino++) { for (int ino = beg_img_no; ino < end_img_no; ino++) {
cv::Mat srcimg; cv::Mat srcimg;
img_list[indices[ino]].copyTo(srcimg); img_list[indices[ino]].copyTo(srcimg);
cv::Mat resize_img; cv::Mat resize_img;
this->resize_op_.Run(srcimg, resize_img, max_wh_ratio, this->resize_op_.Run(srcimg, resize_img, max_wh_ratio,
this->use_tensorrt_); this->use_tensorrt_, this->rec_image_shape_);
this->normalize_op_.Run(&resize_img, this->mean_, this->scale_, this->normalize_op_.Run(&resize_img, this->mean_, this->scale_,
this->is_scale_); this->is_scale_);
norm_img_batch.push_back(resize_img); norm_img_batch.push_back(resize_img);
batch_width = max(resize_img.cols, batch_width); batch_width = max(resize_img.cols, batch_width);
} }
std::vector<float> input(batch_num * 3 * 32 * batch_width, 0.0f); std::vector<float> input(batch_num * 3 * imgH * batch_width, 0.0f);
this->permute_op_.Run(norm_img_batch, input.data()); this->permute_op_.Run(norm_img_batch, input.data());
auto preprocess_end = std::chrono::steady_clock::now(); auto preprocess_end = std::chrono::steady_clock::now();
preprocess_diff += preprocess_end - preprocess_start; preprocess_diff += preprocess_end - preprocess_start;
// Inference. // Inference.
auto input_names = this->predictor_->GetInputNames(); auto input_names = this->predictor_->GetInputNames();
auto input_t = this->predictor_->GetInputHandle(input_names[0]); auto input_t = this->predictor_->GetInputHandle(input_names[0]);
input_t->Reshape({batch_num, 3, 32, batch_width}); input_t->Reshape({batch_num, 3, imgH, batch_width});
auto inference_start = std::chrono::steady_clock::now(); auto inference_start = std::chrono::steady_clock::now();
input_t->CopyFromCpu(input.data()); input_t->CopyFromCpu(input.data());
this->predictor_->Run(); this->predictor_->Run();
...@@ -142,13 +144,14 @@ void CRNNRecognizer::LoadModel(const std::string &model_dir) { ...@@ -142,13 +144,14 @@ void CRNNRecognizer::LoadModel(const std::string &model_dir) {
precision = paddle_infer::Config::Precision::kInt8; precision = paddle_infer::Config::Precision::kInt8;
} }
config.EnableTensorRtEngine(1 << 20, 10, 3, precision, false, false); config.EnableTensorRtEngine(1 << 20, 10, 3, precision, false, false);
int imgH = this->rec_image_shape_[1];
int imgW = this->rec_image_shape_[2];
std::map<std::string, std::vector<int>> min_input_shape = { std::map<std::string, std::vector<int>> min_input_shape = {
{"x", {1, 3, 32, 10}}, {"lstm_0.tmp_0", {10, 1, 96}}}; {"x", {1, 3, imgH, 10}}, {"lstm_0.tmp_0", {10, 1, 96}}};
std::map<std::string, std::vector<int>> max_input_shape = { std::map<std::string, std::vector<int>> max_input_shape = {
{"x", {1, 3, 32, 2000}}, {"lstm_0.tmp_0", {1000, 1, 96}}}; {"x", {1, 3, imgH, 2000}}, {"lstm_0.tmp_0", {1000, 1, 96}}};
std::map<std::string, std::vector<int>> opt_input_shape = { std::map<std::string, std::vector<int>> opt_input_shape = {
{"x", {1, 3, 32, 320}}, {"lstm_0.tmp_0", {25, 1, 96}}}; {"x", {1, 3, imgH, imgW}}, {"lstm_0.tmp_0", {25, 1, 96}}};
config.SetTRTDynamicShapeInfo(min_input_shape, max_input_shape, config.SetTRTDynamicShapeInfo(min_input_shape, max_input_shape,
opt_input_shape); opt_input_shape);
......
deploy/cpp_infer/src/paddleocr.cpp
View file @ aa59fca5
...@@ -17,11 +17,9 @@ ...@@ -17,11 +17,9 @@
#include "auto_log/autolog.h" #include "auto_log/autolog.h"
#include <numeric> #include <numeric>
#include <sys/stat.h>
namespace PaddleOCR { namespace PaddleOCR {
PaddleOCR::PaddleOCR() { PPOCR::PPOCR() {
if (FLAGS_det) { if (FLAGS_det) {
this->detector_ = new DBDetector( this->detector_ = new DBDetector(
FLAGS_det_model_dir, FLAGS_use_gpu, FLAGS_gpu_id, FLAGS_gpu_mem, FLAGS_det_model_dir, FLAGS_use_gpu, FLAGS_gpu_id, FLAGS_gpu_mem,
...@@ -41,12 +39,13 @@ PaddleOCR::PaddleOCR() { ...@@ -41,12 +39,13 @@ PaddleOCR::PaddleOCR() {
this->recognizer_ = new CRNNRecognizer( this->recognizer_ = new CRNNRecognizer(
FLAGS_rec_model_dir, FLAGS_use_gpu, FLAGS_gpu_id, FLAGS_gpu_mem, FLAGS_rec_model_dir, FLAGS_use_gpu, FLAGS_gpu_id, FLAGS_gpu_mem,
FLAGS_cpu_threads, FLAGS_enable_mkldnn, FLAGS_rec_char_dict_path, FLAGS_cpu_threads, FLAGS_enable_mkldnn, FLAGS_rec_char_dict_path,
FLAGS_use_tensorrt, FLAGS_precision, FLAGS_rec_batch_num); FLAGS_use_tensorrt, FLAGS_precision, FLAGS_rec_batch_num,
FLAGS_rec_img_h, FLAGS_rec_img_w);
} }
}; };
void PaddleOCR::det(cv::Mat img, std::vector<OCRPredictResult> &ocr_results, void PPOCR::det(cv::Mat img, std::vector<OCRPredictResult> &ocr_results,
std::vector<double> &times) { std::vector<double> &times) {
std::vector<std::vector<std::vector<int>>> boxes; std::vector<std::vector<std::vector<int>>> boxes;
std::vector<double> det_times; std::vector<double> det_times;
...@@ -63,9 +62,9 @@ void PaddleOCR::det(cv::Mat img, std::vector<OCRPredictResult> &ocr_results, ...@@ -63,9 +62,9 @@ void PaddleOCR::det(cv::Mat img, std::vector<OCRPredictResult> &ocr_results,
times[2] += det_times[2]; times[2] += det_times[2];
} }
void PaddleOCR::rec(std::vector<cv::Mat> img_list, void PPOCR::rec(std::vector<cv::Mat> img_list,
std::vector<OCRPredictResult> &ocr_results, std::vector<OCRPredictResult> &ocr_results,
std::vector<double> &times) { std::vector<double> &times) {
std::vector<std::string> rec_texts(img_list.size(), ""); std::vector<std::string> rec_texts(img_list.size(), "");
std::vector<float> rec_text_scores(img_list.size(), 0); std::vector<float> rec_text_scores(img_list.size(), 0);
std::vector<double> rec_times; std::vector<double> rec_times;
...@@ -80,9 +79,9 @@ void PaddleOCR::rec(std::vector<cv::Mat> img_list, ...@@ -80,9 +79,9 @@ void PaddleOCR::rec(std::vector<cv::Mat> img_list,
times[2] += rec_times[2]; times[2] += rec_times[2];
} }
void PaddleOCR::cls(std::vector<cv::Mat> img_list, void PPOCR::cls(std::vector<cv::Mat> img_list,
std::vector<OCRPredictResult> &ocr_results, std::vector<OCRPredictResult> &ocr_results,
std::vector<double> &times) { std::vector<double> &times) {
std::vector<int> cls_labels(img_list.size(), 0); std::vector<int> cls_labels(img_list.size(), 0);
std::vector<float> cls_scores(img_list.size(), 0); std::vector<float> cls_scores(img_list.size(), 0);
std::vector<double> cls_times; std::vector<double> cls_times;
...@@ -98,8 +97,8 @@ void PaddleOCR::cls(std::vector<cv::Mat> img_list, ...@@ -98,8 +97,8 @@ void PaddleOCR::cls(std::vector<cv::Mat> img_list,
} }
std::vector<std::vector<OCRPredictResult>> std::vector<std::vector<OCRPredictResult>>
PaddleOCR::ocr(std::vector<cv::String> cv_all_img_names, bool det, bool rec, PPOCR::ocr(std::vector<cv::String> cv_all_img_names, bool det, bool rec,
bool cls) { bool cls) {
std::vector<double> time_info_det = {0, 0, 0}; std::vector<double> time_info_det = {0, 0, 0};
std::vector<double> time_info_rec = {0, 0, 0}; std::vector<double> time_info_rec = {0, 0, 0};
std::vector<double> time_info_cls = {0, 0, 0}; std::vector<double> time_info_cls = {0, 0, 0};
...@@ -139,7 +138,7 @@ PaddleOCR::ocr(std::vector<cv::String> cv_all_img_names, bool det, bool rec, ...@@ -139,7 +138,7 @@ PaddleOCR::ocr(std::vector<cv::String> cv_all_img_names, bool det, bool rec,
} }
} else { } else {
if (!Utility::PathExists(FLAGS_output) && FLAGS_det) { if (!Utility::PathExists(FLAGS_output) && FLAGS_det) {
mkdir(FLAGS_output.c_str(), 0777); Utility::CreateDir(FLAGS_output);
} }
for (int i = 0; i < cv_all_img_names.size(); ++i) { for (int i = 0; i < cv_all_img_names.size(); ++i) {
...@@ -188,9 +187,8 @@ PaddleOCR::ocr(std::vector<cv::String> cv_all_img_names, bool det, bool rec, ...@@ -188,9 +187,8 @@ PaddleOCR::ocr(std::vector<cv::String> cv_all_img_names, bool det, bool rec,
return ocr_results; return ocr_results;
} // namespace PaddleOCR } // namespace PaddleOCR
void PaddleOCR::log(std::vector<double> &det_times, void PPOCR::log(std::vector<double> &det_times, std::vector<double> &rec_times,
std::vector<double> &rec_times, std::vector<double> &cls_times, int img_num) {
std::vector<double> &cls_times, int img_num) {
if (det_times[0] + det_times[1] + det_times[2] > 0) { if (det_times[0] + det_times[1] + det_times[2] > 0) {
AutoLogger autolog_det("ocr_det", FLAGS_use_gpu, FLAGS_use_tensorrt, AutoLogger autolog_det("ocr_det", FLAGS_use_gpu, FLAGS_use_tensorrt,
FLAGS_enable_mkldnn, FLAGS_cpu_threads, 1, "dynamic", FLAGS_enable_mkldnn, FLAGS_cpu_threads, 1, "dynamic",
...@@ -212,7 +210,7 @@ void PaddleOCR::log(std::vector<double> &det_times, ...@@ -212,7 +210,7 @@ void PaddleOCR::log(std::vector<double> &det_times,
autolog_cls.report(); autolog_cls.report();
} }
} }
PaddleOCR::~PaddleOCR() { PPOCR::~PPOCR() {
if (this->detector_ != nullptr) { if (this->detector_ != nullptr) {
delete this->detector_; delete this->detector_;
} }
......
deploy/cpp_infer/src/preprocess_op.cpp
View file @ aa59fca5
...@@ -41,16 +41,17 @@ void Permute::Run(const cv::Mat *im, float *data) { ...@@ -41,16 +41,17 @@ void Permute::Run(const cv::Mat *im, float *data) {
} }
void PermuteBatch::Run(const std::vector<cv::Mat> imgs, float *data) { void PermuteBatch::Run(const std::vector<cv::Mat> imgs, float *data) {
for (int j = 0; j < imgs.size(); j ++){ for (int j = 0; j < imgs.size(); j++) {
int rh = imgs[j].rows; int rh = imgs[j].rows;
int rw = imgs[j].cols; int rw = imgs[j].cols;
int rc = imgs[j].channels(); int rc = imgs[j].channels();
for (int i = 0; i < rc; ++i) { for (int i = 0; i < rc; ++i) {
cv::extractChannel(imgs[j], cv::Mat(rh, rw, CV_32FC1, data + (j * rc + i) * rh * rw), i); cv::extractChannel(
} imgs[j], cv::Mat(rh, rw, CV_32FC1, data + (j * rc + i) * rh * rw), i);
} }
}
} }
void Normalize::Run(cv::Mat *im, const std::vector<float> &mean, void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
const std::vector<float> &scale, const bool is_scale) { const std::vector<float> &scale, const bool is_scale) {
double e = 1.0; double e = 1.0;
...@@ -101,8 +102,8 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio, ...@@ -101,8 +102,8 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
imgC = rec_image_shape[0]; imgC = rec_image_shape[0];
imgH = rec_image_shape[1]; imgH = rec_image_shape[1];
imgW = rec_image_shape[2]; imgW = rec_image_shape[2];
imgW = int(32 * wh_ratio); imgW = int(imgH * wh_ratio);
float ratio = float(img.cols) / float(img.rows); float ratio = float(img.cols) / float(img.rows);
int resize_w, resize_h; int resize_w, resize_h;
...@@ -111,7 +112,7 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio, ...@@ -111,7 +112,7 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
resize_w = imgW; resize_w = imgW;
else else
resize_w = int(ceilf(imgH * ratio)); resize_w = int(ceilf(imgH * ratio));
cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f, cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
cv::INTER_LINEAR); cv::INTER_LINEAR);
cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
......
deploy/cpp_infer/src/utility.cpp
View file @ aa59fca5
...@@ -16,10 +16,15 @@ ...@@ -16,10 +16,15 @@
#include <include/utility.h> #include <include/utility.h>
#include <iostream> #include <iostream>
#include <ostream> #include <ostream>
#include <sys/stat.h>
#include <sys/types.h>
#include <vector> #include <vector>
#ifdef _WIN32
#include <direct.h>
#else
#include <sys/stat.h>
#endif
namespace PaddleOCR { namespace PaddleOCR {
std::vector<std::string> Utility::ReadDict(const std::string &path) { std::vector<std::string> Utility::ReadDict(const std::string &path) {
...@@ -206,6 +211,14 @@ bool Utility::PathExists(const std::string &path) { ...@@ -206,6 +211,14 @@ bool Utility::PathExists(const std::string &path) {
#endif // !_WIN32 #endif // !_WIN32
} }
void Utility::CreateDir(const std::string &path) {
#ifdef _WIN32
_mkdir(path.c_str());
#else
mkdir(path.c_str(), 0777);
#endif // !_WIN32
}
void Utility::print_result(const std::vector<OCRPredictResult> &ocr_result) { void Utility::print_result(const std::vector<OCRPredictResult> &ocr_result) {
for (int i = 0; i < ocr_result.size(); i++) { for (int i = 0; i < ocr_result.size(); i++) {
std::cout << i << "\t"; std::cout << i << "\t";
......
deploy/lite/readme.md
View file @ aa59fca5
This diff is collapsed.
deploy/lite/readme_ch.md 0 → 100644
View file @ aa59fca5
- [端侧部署](#端侧部署)
- [1. 准备环境](#1-准备环境)
- [运行准备](#运行准备)
- [1.1 准备交叉编译环境](#11-准备交叉编译环境)
- [1.2 准备预测库](#12-准备预测库)
- [2 开始运行](#2-开始运行)
- [2.1 模型优化](#21-模型优化)
- [2.2 与手机联调](#22-与手机联调)
- [注意：](#注意)
- [FAQ](#faq)
# 端侧部署
本教程将介绍基于[Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) 在移动端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。
## 1. 准备环境
### 运行准备
- 电脑（编译Paddle Lite）
- 安卓手机（armv7或armv8）
### 1.1 准备交叉编译环境
交叉编译环境用于编译 Paddle Lite 和 PaddleOCR 的C++ demo。
支持多种开发环境，不同开发环境的编译流程请参考对应文档。
1. [Docker](https://paddle-lite.readthedocs.io/zh/latest/source_compile/compile_env.html#docker)
2. [Linux](https://paddle-lite.readthedocs.io/zh/latest/source_compile/compile_env.html#linux)
3. [MAC OS](https://paddle-lite.readthedocs.io/zh/latest/source_compile/compile_env.html#mac-os)
### 1.2 准备预测库
预测库有两种获取方式：
- 1. 直接下载，预测库下载链接如下：
| 平台 | 预测库下载链接 |
|---|---|
|Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
|IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
注：1. 上述预测库为PaddleLite 2.10分支编译得到，有关PaddleLite 2.10 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10) 。
- 2. [推荐]编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
```
git clone https://github.com/PaddlePaddle/Paddle-Lite.git
cd Paddle-Lite
# 切换到Paddle-Lite release/v2.10 稳定分支
git checkout release/v2.10
./lite/tools/build_android.sh --arch=armv8 --with_cv=ON --with_extra=ON
```
注意：编译Paddle-Lite获得预测库时，需要打开`--with_cv=ON --with_extra=ON`两个选项，`--arch`表示`arm`版本，这里指定为armv8，
更多编译命令
介绍请参考 [链接](https://paddle-lite.readthedocs.io/zh/release-v2.10_a/source_compile/linux_x86_compile_android.html) 。
直接下载预测库并解压后，可以得到`inference_lite_lib.android.armv8/`文件夹，通过编译Paddle-Lite得到的预测库位于
`Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/`文件夹下。
预测库的文件目录如下：
```
inference_lite_lib.android.armv8/
|-- cxx C++ 预测库和头文件
| |-- include C++ 头文件
| | |-- paddle_api.h
| | |-- paddle_image_preprocess.h
| | |-- paddle_lite_factory_helper.h
| | |-- paddle_place.h
| | |-- paddle_use_kernels.h
| | |-- paddle_use_ops.h
| | `-- paddle_use_passes.h
| `-- lib C++预测库
| |-- libpaddle_api_light_bundled.a C++静态库
| `-- libpaddle_light_api_shared.so C++动态库
|-- java Java预测库
| |-- jar
| | `-- PaddlePredictor.jar
| |-- so
| | `-- libpaddle_lite_jni.so
| `-- src
|-- demo C++和Java示例代码
| |-- cxx C++ 预测库demo
| `-- java Java 预测库demo
```
## 2 开始运行
### 2.1 模型优化
Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括量化、子图融合、混合调度、Kernel优选等方法，使用Paddle-lite的opt工具可以自动
对inference模型进行优化，优化后的模型更轻量，模型运行速度更快。
如果已经准备好了 `.nb` 结尾的模型文件，可以跳过此步骤。
下述表格中也提供了一系列中文移动端模型：
|模型版本|模型简介|模型大小|检测模型|文本方向分类模型|识别模型|Paddle-Lite版本|
|---|---|---|---|---|---|---|
|PP-OCRv2|蒸馏版超轻量中文OCR移动端模型|11M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
|PP-OCRv2(slim)|蒸馏版超轻量中文OCR移动端模型|4.6M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
如果直接使用上述表格中的模型进行部署，可略过下述步骤，直接阅读 [2.2节](#2.2与手机联调)。
如果要部署的模型不在上述表格中，则需要按照如下步骤获得优化后的模型。
模型优化需要Paddle-Lite的opt可执行文件，可以通过编译Paddle-Lite源码获得，编译步骤如下：
```
# 如果准备环境时已经clone了Paddle-Lite，则不用重新clone Paddle-Lite
git clone https://github.com/PaddlePaddle/Paddle-Lite.git
cd Paddle-Lite
git checkout release/v2.10
# 启动编译
./lite/tools/build.sh build_optimize_tool
```
编译完成后，opt文件位于`build.opt/lite/api/`下，可通过如下方式查看opt的运行选项和使用方式；
```
cd build.opt/lite/api/
./opt
```
|选项|说明|
|---|---|
|--model_dir|待优化的PaddlePaddle模型（非combined形式）的路径|
|--model_file|待优化的PaddlePaddle模型（combined形式）的网络结构文件路径|
|--param_file|待优化的PaddlePaddle模型（combined形式）的权重文件路径|
|--optimize_out_type|输出模型类型，目前支持两种类型：protobuf和naive_buffer，其中naive_buffer是一种更轻量级的序列化/反序列化实现。若您需要在mobile端执行模型预测，请将此选项设置为naive_buffer。默认为protobuf|
|--optimize_out|优化模型的输出路径|
|--valid_targets|指定模型可执行的backend，默认为arm。目前可支持x86、arm、opencl、npu、xpu，可以同时指定多个backend(以空格分隔)，Model Optimize Tool将会自动选择最佳方式。如果需要支持华为NPU（Kirin 810/990 Soc搭载的达芬奇架构NPU），应当设置为npu, arm|
|--record_tailoring_info|当使用 根据模型裁剪库文件 功能时，则设置该选项为true，以记录优化后模型含有的kernel和OP信息，默认为false|
`--model_dir`适用于待优化的模型是非combined方式，PaddleOCR的inference模型是combined方式，即模型结构和模型参数使用单独一个文件存储。
下面以PaddleOCR的超轻量中文模型为例，介绍使用编译好的opt文件完成inference模型到Paddle-Lite优化模型的转换。
```
# 【推荐】 下载 PP-OCRv2版本的中英文 inference模型
wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_quant_infer.tar && tar xf ch_PP-OCRv2_det_slim_quant_infer.tar
wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar && tar xf ch_PP-OCRv2_rec_slim_quant_infer.tar
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf ch_ppocr_mobile_v2.0_cls_slim_infer.tar
# 转换检测模型
./opt --model_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdmodel --param_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdiparams --optimize_out=./ch_PP-OCRv2_det_slim_opt --valid_targets=arm --optimize_out_type=naive_buffer
# 转换识别模型
./opt --model_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdmodel --param_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdiparams --optimize_out=./ch_PP-OCRv2_rec_slim_opt --valid_targets=arm --optimize_out_type=naive_buffer
# 转换方向分类器模型
./opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm --optimize_out_type=naive_buffer
```
转换成功后，inference模型目录下会多出`.nb`结尾的文件，即是转换成功的模型文件。
注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt 工具的输入模型是paddle保存的inference模型
<a name="2.2与手机联调"></a>
### 2.2 与手机联调
首先需要进行一些准备工作。
1. 准备一台arm8的安卓手机，如果编译的预测库和opt文件是armv7，则需要arm7的手机，并修改Makefile中`ARM_ABI = arm7`。
2. 打开手机的USB调试选项，选择文件传输模式，连接电脑。
3. 电脑上安装adb工具，用于调试。 adb安装方式如下：
3.1. MAC电脑安装ADB:
```
brew cask install android-platform-tools
```
3.2. Linux安装ADB
```
sudo apt update
sudo apt install -y wget adb
```
3.3. Window安装ADB
win上安装需要去谷歌的安卓平台下载adb软件包进行安装：[链接](https://developer.android.com/studio)
打开终端，手机连接电脑，在终端中输入
```
adb devices
```
如果有device输出，则表示安装成功。
```
List of devices attached
744be294 device
```
4. 准备优化后的模型、预测库文件、测试图像和使用的字典文件。
```
git clone https://github.com/PaddlePaddle/PaddleOCR.git
cd PaddleOCR/deploy/lite/
# 运行prepare.sh，准备预测库文件、测试图像和使用的字典文件，并放置在预测库中的demo/cxx/ocr文件夹下
sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
# 进入OCR demo的工作目录
cd /{lite prediction library path}/inference_lite_lib.android.armv8/
cd demo/cxx/ocr/
# 将C++预测动态库so文件复制到debug文件夹中
cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
```
准备测试图像，以`PaddleOCR/doc/imgs/11.jpg`为例，将测试的图像复制到`demo/cxx/ocr/debug/`文件夹下。
准备lite opt工具优化后的模型文件，比如使用`ch_PP-OCRv2_det_slim_opt.ch_PP-OCRv2_rec_slim_rec.nb, ch_ppocr_mobile_v2.0_cls_slim_opt.nb`，模型文件放置在`demo/cxx/ocr/debug/`文件夹下。
执行完成后，ocr文件夹下将有如下文件格式：
```
demo/cxx/ocr/
|-- debug/
| |--ch_PP-OCRv2_det_slim_opt.nb 优化后的检测模型文件
| |--ch_PP-OCRv2_rec_slim_opt.nb 优化后的识别模型文件
| |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb 优化后的文字方向分类器模型文件
| |--11.jpg 待测试图像
| |--ppocr_keys_v1.txt 中文字典文件
| |--libpaddle_light_api_shared.so C++预测库文件
| |--config.txt 超参数配置
|-- config.txt 超参数配置
|-- cls_process.cc 方向分类器的预处理和后处理文件
|-- cls_process.h
|-- crnn_process.cc 识别模型CRNN的预处理和后处理文件
|-- crnn_process.h
|-- db_post_process.cc 检测模型DB的后处理文件
|-- db_post_process.h
|-- Makefile 编译文件
|-- ocr_db_crnn.cc C++预测源文件
```
#### 注意：
1. ppocr_keys_v1.txt是中文字典文件，如果使用的 nb 模型是英文数字或其他语言的模型，需要更换为对应语言的字典。
PaddleOCR 在ppocr/utils/下存放了多种字典，包括：
```
dict/french_dict.txt # 法语字典
dict/german_dict.txt # 德语字典
ic15_dict.txt # 英文字典
dict/japan_dict.txt # 日语字典
dict/korean_dict.txt # 韩语字典
ppocr_keys_v1.txt # 中文字典
...
```
2. `config.txt` 包含了检测器、分类器的超参数，如下：
```
max_side_len 960 # 输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960
det_db_thresh 0.3 # 用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不明显
det_db_box_thresh 0.5 # DB后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小
det_db_unclip_ratio 1.6 # 表示文本框的紧致程度，越小则文本框更靠近文本
use_direction_classify 0 # 是否使用方向分类器，0表示不使用，1表示使用
```
5. 启动调试
上述步骤完成后就可以使用adb将文件push到手机上运行，步骤如下：
```
# 执行编译，得到可执行文件ocr_db_crnn, 第一次执行此命令会下载opencv等依赖库，下载完成后，需要再执行一次
make -j
# 将编译的可执行文件移动到debug文件夹中
mv ocr_db_crnn ./debug/
# 将debug文件夹push到手机上
adb push debug /data/local/tmp/
adb shell
cd /data/local/tmp/debug
export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
# 开始使用，ocr_db_crnn可执行文件的使用方式为:
# ./ocr_db_crnn 检测模型文件 方向分类器模型文件 识别模型文件 测试图像路径 字典文件路径
./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb ch_PP-OCRv2_rec_slim_opt.nb ch_ppocr_mobile_v2.0_cls_slim_opt.nb ./11.jpg ppocr_keys_v1.txt
```
如果对代码做了修改，则需要重新编译并push到手机上。
运行效果如下：
<div align="center">
<img src="imgs/lite_demo.png" width="600">
</div>
## FAQ
Q1：如果想更换模型怎么办，需要重新按照流程走一遍吗？
A1：如果已经走通了上述步骤，更换模型只需要替换 .nb 模型文件即可，同时要注意更新字典
Q2：换一个图测试怎么做？
A2：替换debug下的.jpg测试图像为你想要测试的图像，adb push 到手机上即可
Q3：如何封装到手机APP中？
A3：此demo旨在提供能在手机上运行OCR的核心算法部分，PaddleOCR/deploy/android_demo是将这个demo封装到手机app的示例，供参考
deploy/lite/readme_en.md deleted 100644 → 0
View file @ 12d15752
- [Tutorial of PaddleOCR Mobile deployment](#tutorial-of-paddleocr-mobile-deployment)
- [1. Preparation](#1-preparation)
- [Preparation environment](#preparation-environment)
- [1.1 Prepare the cross-compilation environment](#11-prepare-the-cross-compilation-environment)
- [1.2 Prepare Paddle-Lite library](#12-prepare-paddle-lite-library)
- [2 Run](#2-run)
- [2.1 Inference Model Optimization](#21-inference-model-optimization)
- [2.2 Run optimized model on Phone](#22-run-optimized-model-on-phone)
- [注意：](#注意)
- [FAQ](#faq)
# Tutorial of PaddleOCR Mobile deployment
This tutorial will introduce how to use [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) to deploy PaddleOCR ultra-lightweight Chinese and English detection models on mobile phones.
paddle-lite is a lightweight inference engine for PaddlePaddle. It provides efficient inference capabilities for mobile phones and IoT, and extensively integrates cross-platform hardware to provide lightweight deployment solutions for end-side deployment issues.
## 1. Preparation
### Preparation environment
- Computer (for Compiling Paddle Lite)
- Mobile phone (arm7 or arm8)
### 1.1 Prepare the cross-compilation environment
The cross-compilation environment is used to compile C++ demos of Paddle Lite and PaddleOCR.
Supports multiple development environments.
For the compilation process of different development environments, please refer to the corresponding documents.
1. [Docker](https://paddle-lite.readthedocs.io/zh/latest/source_compile/compile_env.html#docker)
2. [Linux](https://paddle-lite.readthedocs.io/zh/latest/source_compile/compile_env.html#linux)
3. [MAC OS](https://paddle-lite.readthedocs.io/zh/latest/source_compile/compile_env.html#mac-os)
### 1.2 Prepare Paddle-Lite library
There are two ways to obtain the Paddle-Lite library：
- 1. Download directly, the download link of the Paddle-Lite library is as follows：
| Platform | Paddle-Lite library download link |
|---|---|
|Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
|IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.10 branch. For more information about Paddle-Lite 2.10, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10).
- 2. [Recommended] Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
```
git clone https://github.com/PaddlePaddle/Paddle-Lite.git
cd Paddle-Lite
# Switch to Paddle-Lite release/v2.10 stable branch
git checkout release/v2.10
./lite/tools/build_android.sh --arch=armv8 --with_cv=ON --with_extra=ON
```
Note: When compiling Paddle-Lite to obtain the Paddle-Lite library, you need to turn on the two options `--with_cv=ON --with_extra=ON`, `--arch` means the `arm` version, here is designated as armv8,
More compilation commands refer to the introduction [link](https://paddle-lite.readthedocs.io/zh/release-v2.10_a/source_compile/linux_x86_compile_android.html) 。
After directly downloading the Paddle-Lite library and decompressing it, you can get the `inference_lite_lib.android.armv8/` folder, and the Paddle-Lite library obtained by compiling Paddle-Lite is located
`Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/` folder.
The structure of the prediction library is as follows:
```
inference_lite_lib.android.armv8/
|-- cxx C++ prebuild library
| |-- include C++
| | |-- paddle_api.h
| | |-- paddle_image_preprocess.h
| | |-- paddle_lite_factory_helper.h
| | |-- paddle_place.h
| | |-- paddle_use_kernels.h
| | |-- paddle_use_ops.h
| | `-- paddle_use_passes.h
| `-- lib C++ library
| |-- libpaddle_api_light_bundled.a C++ static library
| `-- libpaddle_light_api_shared.so C++ dynamic library
|-- java Java library
| |-- jar
| | `-- PaddlePredictor.jar
| |-- so
| | `-- libpaddle_lite_jni.so
| `-- src
|-- demo C++ and Java demo
| |-- cxx C++ demo
| `-- java Java demo
```
## 2 Run
### 2.1 Inference Model Optimization
Paddle Lite provides a variety of strategies to automatically optimize the original training model, including quantization, sub-graph fusion, hybrid scheduling, Kernel optimization and so on. In order to make the optimization process more convenient and easy to use, Paddle Lite provide opt tools to automatically complete the optimization steps and output a lightweight, optimal executable model.
If you have prepared the model file ending in .nb, you can skip this step.
The following table also provides a series of models that can be deployed on mobile phones to recognize Chinese. You can directly download the optimized model.
|Version|Introduction|Model size|Detection model|Text Direction model|Recognition model|Paddle-Lite branch|
|---|---|---|---|---|---|---|
|PP-OCRv2|extra-lightweight chinese OCR optimized model|11M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
|PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.6M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
If you directly use the model in the above table for deployment, you can skip the following steps and directly read [Section 2.2](#2.2-Run-optimized-model-on-Phone).
If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
The `opt` tool can be obtained by compiling Paddle Lite.
```
git clone https://github.com/PaddlePaddle/Paddle-Lite.git
cd Paddle-Lite
git checkout release/v2.10
./lite/tools/build.sh build_optimize_tool
```
After the compilation is complete, the opt file is located under build.opt/lite/api/, You can view the operating options and usage of opt in the following ways:
```
cd build.opt/lite/api/
./opt
```
|Options|Description|
|---|---|
|--model_dir|The path of the PaddlePaddle model to be optimized (non-combined form)|
|--model_file|The network structure file path of the PaddlePaddle model (combined form) to be optimized|
|--param_file|The weight file path of the PaddlePaddle model (combined form) to be optimized|
|--optimize_out_type|Output model type, currently supports two types: protobuf and naive_buffer, among which naive_buffer is a more lightweight serialization/deserialization implementation. If you need to perform model prediction on the mobile side, please set this option to naive_buffer. The default is protobuf|
|--optimize_out|The output path of the optimized model|
|--valid_targets|The executable backend of the model, the default is arm. Currently it supports x86, arm, opencl, npu, xpu, multiple backends can be specified at the same time (separated by spaces), and Model Optimize Tool will automatically select the best method. If you need to support Huawei NPU (DaVinci architecture NPU equipped with Kirin 810/990 Soc), it should be set to npu, arm|
|--record_tailoring_info|When using the function of cutting library files according to the model, set this option to true to record the kernel and OP information contained in the optimized model. The default is false|
`--model_dir` is suitable for the non-combined mode of the model to be optimized, and the inference model of PaddleOCR is the combined mode, that is, the model structure and model parameters are stored in a single file.
The following takes the ultra-lightweight Chinese model of PaddleOCR as an example to introduce the use of the compiled opt file to complete the conversion of the inference model to the Paddle-Lite optimized model
```
# 【[Recommendation] Download the Chinese and English inference model of PP-OCRv2
wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_quant_infer.tar && tar xf ch_PP-OCRv2_det_slim_quant_infer.tar
wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar && tar xf ch_PP-OCRv2_rec_slim_quant_infer.tar
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf ch_ppocr_mobile_v2.0_cls_slim_infer.tar
# Convert detection model
./opt --model_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdmodel --param_file=./ch_PP-OCRv2_det_slim_quant_infer/inference.pdiparams --optimize_out=./ch_PP-OCRv2_det_slim_opt --valid_targets=arm --optimize_out_type=naive_buffer
# Convert recognition model
./opt --model_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdmodel --param_file=./ch_PP-OCRv2_rec_slim_quant_infer/inference.pdiparams --optimize_out=./ch_PP-OCRv2_rec_slim_opt --valid_targets=arm --optimize_out_type=naive_buffer
# Convert angle classifier model
./opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm --optimize_out_type=naive_buffer
```
After the conversion is successful, there will be more files ending with `.nb` in the inference model directory, which is the successfully converted model file.
<a name="2.2-Run-optimized-model-on-Phone"></a>
### 2.2 Run optimized model on Phone
Some preparatory work is required first.
1. Prepare an Android phone with arm8. If the compiled prediction library and opt file are armv7, you need an arm7 phone and modify ARM_ABI = arm7 in the Makefile.
2. Make sure the phone is connected to the computer, open the USB debugging option of the phone, and select the file transfer mode.
3. Install the adb tool on the computer.
3.1. Install ADB for MAC:
```
brew cask install android-platform-tools
```
3.2. Install ADB for Linux
```
sudo apt update
sudo apt install -y wget adb
```
3.3. Install ADB for windows
To install on win, you need to go to Google's Android platform to download the adb package for installation：[link](https://developer.android.com/studio)
Verify whether adb is installed successfully
```
adb devices
```
If there is device output, it means the installation is successful。
```
List of devices attached
744be294 device
```
4. Prepare optimized models, prediction library files, test images and dictionary files used.
```
git clone https://github.com/PaddlePaddle/PaddleOCR.git
cd PaddleOCR/deploy/lite/
# run prepare.sh
sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
#
cd /{lite prediction library path}/inference_lite_lib.android.armv8/
cd demo/cxx/ocr/
# copy paddle-lite C++ .so file to debug/ directory
cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
cd inference_lite_lib.android.armv8/demo/cxx/ocr/
cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
```
Prepare the test image, taking PaddleOCR/doc/imgs/11.jpg as an example, copy the image file to the demo/cxx/ocr/debug/ folder. Prepare the model files optimized by the lite opt tool, ch_det_mv3_db_opt.nb, ch_rec_mv3_crnn_opt.nb, and place them under the demo/cxx/ocr/debug/ folder.
The structure of the OCR demo is as follows after the above command is executed:
```
demo/cxx/ocr/
|-- debug/
| |--ch_PP-OCRv2_det_slim_opt.nb Detection model
| |--ch_PP-OCRv2_rec_slim_opt.nb Recognition model
| |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb Text direction classification model
| |--11.jpg Image for OCR
| |--ppocr_keys_v1.txt Dictionary file
| |--libpaddle_light_api_shared.so C++ .so file
| |--config.txt Config file
|-- config.txt Config file
|-- cls_process.cc Pre-processing and post-processing files for the angle classifier
|-- cls_process.h
|-- crnn_process.cc Pre-processing and post-processing files for the CRNN model
|-- crnn_process.h
|-- db_post_process.cc Pre-processing and post-processing files for the DB model
|-- db_post_process.h
|-- Makefile
|-- ocr_db_crnn.cc C++ main code
```
#### 注意：
1. `ppocr_keys_v1.txt` is a Chinese dictionary file. If the nb model is used for English recognition or other language recognition, dictionary file should be replaced with a dictionary of the corresponding language. PaddleOCR provides a variety of dictionaries under ppocr/utils/, including:
```
dict/french_dict.txt # french
dict/german_dict.txt # german
ic15_dict.txt # english
dict/japan_dict.txt # japan
dict/korean_dict.txt # korean
ppocr_keys_v1.txt # chinese
```
2. `config.txt` of the detector and classifier, as shown below:
```
max_side_len 960 # Limit the maximum image height and width to 960
det_db_thresh 0.3 # Used to filter the binarized image of DB prediction, setting 0.-0.3 has no obvious effect on the result
det_db_box_thresh 0.5 # DDB post-processing filter box threshold, if there is a missing box detected, it can be reduced as appropriate
det_db_unclip_ratio 1.6 # Indicates the compactness of the text box, the smaller the value, the closer the text box to the text
use_direction_classify 0 # Whether to use the direction classifier, 0 means not to use, 1 means to use
```
5. Run Model on phone
After the above steps are completed, you can use adb to push the file to the phone to run, the steps are as follows:
```
# Execute the compilation and get the executable file ocr_db_crnn
# The first execution of this command will download dependent libraries such as opencv. After the download is complete, you need to execute it again
make -j
# Move the compiled executable file to the debug folder
mv ocr_db_crnn ./debug/
# Push the debug folder to the phone
adb push debug /data/local/tmp/
adb shell
cd /data/local/tmp/debug
export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
# The use of ocr_db_crnn is:
# ./ocr_db_crnn Detection model file Orientation classifier model file Recognition model file Test image path Dictionary file path
./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb ch_PP-OCRv2_rec_slim_opt.nb ch_ppocr_mobile_v2.0_cls_opt.nb ./11.jpg ppocr_keys_v1.txt
```
If you modify the code, you need to recompile and push to the phone.
The outputs are as follows:
<div align="center">
<img src="imgs/lite_demo.png" width="600">
</div>
## FAQ
Q1: What if I want to change the model, do I need to run it again according to the process?
A1: If you have performed the above steps, you only need to replace the .nb model file to complete the model replacement.
Q2: How to test with another picture?
A2: Replace the .jpg test image under ./debug with the image you want to test, and run adb push to push new image to the phone.
Q3: How to package it into the mobile APP?
A3: This demo aims to provide the core algorithm part that can run OCR on mobile phones. Further, PaddleOCR/deploy/android_demo is an example of encapsulating this demo into a mobile app for reference.
deploy/paddle2onnx/readme_ch.md 0 → 100644
View file @ aa59fca5
# Paddle2ONNX模型转化与预测
本章节介绍 PaddleOCR 模型如何转化为 ONNX 模型，并基于 ONNXRuntime 引擎预测。
## 1. 环境准备
需要准备 PaddleOCR、Paddle2ONNX 模型转化环境，和 ONNXRuntime 预测环境
### PaddleOCR
克隆PaddleOCR的仓库，使用release/2.4分支，并进行安装，由于PaddleOCR仓库比较大，git clone速度比较慢，所以本教程已下载
```
git clone -b release/2.4 https://github.com/PaddlePaddle/PaddleOCR.git
cd PaddleOCR && python3.7 setup.py install
```
### Paddle2ONNX
Paddle2ONNX 支持将 PaddlePaddle 模型格式转化到 ONNX 模型格式，算子目前稳定支持导出 ONNX Opset 9~11，部分Paddle算子支持更低的ONNX Opset转换。
更多细节可参考 [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX/blob/develop/README_zh.md)
- 安装 Paddle2ONNX
```
python3.7 -m pip install paddle2onnx
```
- 安装 ONNXRuntime
```
# 建议安装 1.9.0 版本，可根据环境更换版本号
python3.7 -m pip install onnxruntime==1.9.0
```
## 2. 模型转换
- Paddle 模型下载
有两种方式获取Paddle静态图模型：在 [model_list](../../doc/doc_ch/models_list.md) 中下载PaddleOCR提供的预测模型；
参考[模型导出说明](../../doc/doc_ch/inference.md#训练模型转inference模型)把训练好的权重转为 inference_model。
以 ppocr 中文检测、识别、分类模型为例：
```
wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar
cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && cd ..
wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar
cd ./inference && tar xf ch_PP-OCRv2_rec_infer.tar && cd ..
wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
```
- 模型转换
使用 Paddle2ONNX 将Paddle静态图模型转换为ONNX模型格式：
```
paddle2onnx --model_dir ./inference/ch_PP-OCRv2_det_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_file ./inference/det_onnx/model.onnx \
--opset_version 10 \
--input_shape_dict="{'x':[-1,3,-1,-1]}" \
--enable_onnx_checker True
paddle2onnx --model_dir ./inference/ch_PP-OCRv2_rec_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_file ./inference/rec_onnx/model.onnx \
--opset_version 10 \
--input_shape_dict="{'x':[-1,3,-1,-1]}" \
--enable_onnx_checker True
paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
--model_filename ch_ppocr_mobile_v2.0_cls_infer/inference.pdmodel \
--params_filename ch_ppocr_mobile_v2.0_cls_infer/inference.pdiparams \
--save_file ./inferencecls_onnx/model.onnx \
--opset_version 10 \
--input_shape_dict="{'x':[-1,3,-1,-1]}" \
--enable_onnx_checker True
```
执行完毕后，ONNX 模型会被分别保存在 `./inference/det_onnx/`，`./inference/rec_onnx/`，`./inference/cls_onnx/`路径下
* 注意：对于OCR模型，转化过程中必须采用动态shape的形式，即加入选项--input_shape_dict="{'x': [-1, 3, -1, -1]}"，否则预测结果可能与直接使用Paddle预测有细微不同。
另外，以下几个模型暂不支持转换为 ONNX 模型：
NRTR、SAR、RARE、SRN
## 3. 推理预测
以中文OCR模型为例，使用 ONNXRuntime 预测可执行如下命令：
```
python3.7 tools/infer/predict_system.py --use_gpu=False --use_onnx=True \
--det_model_dir=./inference/det_onnx/model.onnx \
--rec_model_dir=./inference/rec_onnx/model.onnx \
--cls_model_dir=./inference/cls_onnx/model.onnx \
--image_dir=./deploy/lite/imgs/lite_demo.png
```
以中文OCR模型为例，使用 Paddle Inference 预测可执行如下命令：
```
python3.7 tools/infer/predict_system.py --use_gpu=False \
--cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
--rec_model_dir=./inference/ch_PP-OCRv2_rec_infer \
--det_model_dir=./inference/ch_PP-OCRv2_det_infer \
--image_dir=./deploy/lite/imgs/lite_demo.png
```
执行命令后在终端会打印出预测的识别信息，并在 `./inference_results/` 下保存可视化结果。
ONNXRuntime 执行效果：
<div align="center">
<img src="./images/lite_demo_onnx.png" width=800">
</div>
Paddle Inference 执行效果：
<div align="center">
<img src="./images/lite_demo_paddle.png" width=800">
</div>
使用 ONNXRuntime 预测，终端输出：
```
[2022/02/22 17:48:27] root DEBUG: dt_boxes num : 38, elapse : 0.043187856674194336
[2022/02/22 17:48:27] root DEBUG: rec_res num : 38, elapse : 0.592170000076294
[2022/02/22 17:48:27] root DEBUG: 0 Predict time of ./deploy/lite/imgs/lite_demo.png: 0.642s
[2022/02/22 17:48:27] root DEBUG: The, 0.984
[2022/02/22 17:48:27] root DEBUG: visualized, 0.882
[2022/02/22 17:48:27] root DEBUG: etect18片, 0.720
[2022/02/22 17:48:27] root DEBUG: image saved in./vis.jpg, 0.947
[2022/02/22 17:48:27] root DEBUG: 纯臻营养护发素0.993604, 0.996
[2022/02/22 17:48:27] root DEBUG: 产品信息/参数, 0.922
[2022/02/22 17:48:27] root DEBUG: 0.992728, 0.914
[2022/02/22 17:48:27] root DEBUG: （45元／每公斤，100公斤起订）, 0.926
[2022/02/22 17:48:27] root DEBUG: 0.97417, 0.977
[2022/02/22 17:48:27] root DEBUG: 每瓶22元，1000瓶起订）0.993976, 0.962
[2022/02/22 17:48:27] root DEBUG: 【品牌】：代加工方式/0EMODM, 0.945
[2022/02/22 17:48:27] root DEBUG: 0.985133, 0.980
[2022/02/22 17:48:27] root DEBUG: 【品名】：纯臻营养护发素, 0.921
[2022/02/22 17:48:27] root DEBUG: 0.995007, 0.883
[2022/02/22 17:48:27] root DEBUG: 【产品编号】：YM-X-30110.96899, 0.955
[2022/02/22 17:48:27] root DEBUG: 【净含量】：220ml, 0.943
[2022/02/22 17:48:27] root DEBUG: Q.996577, 0.932
[2022/02/22 17:48:27] root DEBUG: 【适用人群】：适合所有肤质, 0.913
[2022/02/22 17:48:27] root DEBUG: 0.995842, 0.969
[2022/02/22 17:48:27] root DEBUG: 【主要成分】：鲸蜡硬脂醇、燕麦B-葡聚, 0.883
[2022/02/22 17:48:27] root DEBUG: 0.961928, 0.964
[2022/02/22 17:48:27] root DEBUG: 10, 0.812
[2022/02/22 17:48:27] root DEBUG: 糖、椰油酰胺丙基甜菜碱、泛醒, 0.866
[2022/02/22 17:48:27] root DEBUG: 0.925898, 0.943
[2022/02/22 17:48:27] root DEBUG: （成品包材）, 0.974
[2022/02/22 17:48:27] root DEBUG: 0.972573, 0.961
[2022/02/22 17:48:27] root DEBUG: 【主要功能】：可紧致头发磷层，从而达到, 0.936
[2022/02/22 17:48:27] root DEBUG: 0.994448, 0.952
[2022/02/22 17:48:27] root DEBUG: 13, 0.998
[2022/02/22 17:48:27] root DEBUG: 即时持久改善头发光泽的效果，给干燥的头, 0.994
[2022/02/22 17:48:27] root DEBUG: 0.990198, 0.975
[2022/02/22 17:48:27] root DEBUG: 14, 0.977
[2022/02/22 17:48:27] root DEBUG: 发足够的滋养, 0.991
[2022/02/22 17:48:27] root DEBUG: 0.997668, 0.918
[2022/02/22 17:48:27] root DEBUG: 花费了0.457335秒, 0.901
[2022/02/22 17:48:27] root DEBUG: The visualized image saved in ./inference_results/lite_demo.png
[2022/02/22 17:48:27] root INFO: The predict total time is 0.7003889083862305
```
使用 Paddle Inference 预测，终端输出：
```
[2022/02/22 17:47:25] root DEBUG: dt_boxes num : 38, elapse : 0.11791276931762695
[2022/02/22 17:47:27] root DEBUG: rec_res num : 38, elapse : 2.6206860542297363
[2022/02/22 17:47:27] root DEBUG: 0 Predict time of ./deploy/lite/imgs/lite_demo.png: 2.746s
[2022/02/22 17:47:27] root DEBUG: The, 0.984
[2022/02/22 17:47:27] root DEBUG: visualized, 0.882
[2022/02/22 17:47:27] root DEBUG: etect18片, 0.720
[2022/02/22 17:47:27] root DEBUG: image saved in./vis.jpg, 0.947
[2022/02/22 17:47:27] root DEBUG: 纯臻营养护发素0.993604, 0.996
[2022/02/22 17:47:27] root DEBUG: 产品信息/参数, 0.922
[2022/02/22 17:47:27] root DEBUG: 0.992728, 0.914
[2022/02/22 17:47:27] root DEBUG: （45元／每公斤，100公斤起订）, 0.926
[2022/02/22 17:47:27] root DEBUG: 0.97417, 0.977
[2022/02/22 17:47:27] root DEBUG: 每瓶22元，1000瓶起订）0.993976, 0.962
[2022/02/22 17:47:27] root DEBUG: 【品牌】：代加工方式/0EMODM, 0.945
[2022/02/22 17:47:27] root DEBUG: 0.985133, 0.980
[2022/02/22 17:47:27] root DEBUG: 【品名】：纯臻营养护发素, 0.921
[2022/02/22 17:47:27] root DEBUG: 0.995007, 0.883
[2022/02/22 17:47:27] root DEBUG: 【产品编号】：YM-X-30110.96899, 0.955
[2022/02/22 17:47:27] root DEBUG: 【净含量】：220ml, 0.943
[2022/02/22 17:47:27] root DEBUG: Q.996577, 0.932
[2022/02/22 17:47:27] root DEBUG: 【适用人群】：适合所有肤质, 0.913
[2022/02/22 17:47:27] root DEBUG: 0.995842, 0.969
[2022/02/22 17:47:27] root DEBUG: 【主要成分】：鲸蜡硬脂醇、燕麦B-葡聚, 0.883
[2022/02/22 17:47:27] root DEBUG: 0.961928, 0.964
[2022/02/22 17:47:27] root DEBUG: 10, 0.812
[2022/02/22 17:47:27] root DEBUG: 糖、椰油酰胺丙基甜菜碱、泛醒, 0.866
[2022/02/22 17:47:27] root DEBUG: 0.925898, 0.943
[2022/02/22 17:47:27] root DEBUG: （成品包材）, 0.974
[2022/02/22 17:47:27] root DEBUG: 0.972573, 0.961
[2022/02/22 17:47:27] root DEBUG: 【主要功能】：可紧致头发磷层，从而达到, 0.936
[2022/02/22 17:47:27] root DEBUG: 0.994448, 0.952
[2022/02/22 17:47:27] root DEBUG: 13, 0.998
[2022/02/22 17:47:27] root DEBUG: 即时持久改善头发光泽的效果，给干燥的头, 0.994
[2022/02/22 17:47:27] root DEBUG: 0.990198, 0.975
[2022/02/22 17:47:27] root DEBUG: 14, 0.977
[2022/02/22 17:47:27] root DEBUG: 发足够的滋养, 0.991
[2022/02/22 17:47:27] root DEBUG: 0.997668, 0.918
[2022/02/22 17:47:27] root DEBUG: 花费了0.457335秒, 0.901
[2022/02/22 17:47:27] root DEBUG: The visualized image saved in ./inference_results/lite_demo.png
[2022/02/22 17:47:27] root INFO: The predict total time is 2.8338775634765625
```
deploy/paddlejs/README.md 0 → 100644
View file @ aa59fca5
English| [简体中文](README_ch.md)
# Paddle.js
[Paddle.js](https://github.com/PaddlePaddle/Paddle.js) is a web project for Baidu PaddlePaddle, which is an open source deep learning framework running in the browser. Paddle.js can either load a pre-trained model, or transforming a model from paddle-hub with model transforming tools provided by Paddle.js. It could run in every browser with WebGL/WebGPU/WebAssembly supported. It could also run in Baidu Smartprogram and WX miniprogram.
- [Online experience](https://paddlejs.baidu.com/ocr)
- [Tutorial](https://github.com/PaddlePaddle/Paddle.js/blob/release/v2.2.3/packages/paddlejs-models/ocr/README_cn.md)
- Visualization：
<div align="center">
<img src="./paddlejs_demo.gif" width="800">
</div>
\ No newline at end of file
deploy/paddlejs/README_ch.md 0 → 100644
View file @ aa59fca5
[English](README.md) | 简体中文
# Paddle.js 网页前端部署
[Paddle.js](https://github.com/PaddlePaddle/Paddle.js) 是百度 PaddlePaddle 的 web 方向子项目，是一个运行在浏览器中的开源深度学习框架。Paddle.js 可以加载提前训练好的 paddle 模型，通过 Paddle.js 的模型转换工具 paddlejs-converter 变成浏览器友好的模型进行在线推理预测使用。目前，Paddle.js 可以在支持 WebGL/WebGPU/WebAssembly 的浏览器中运行，也可以在百度小程序和微信小程序环境下运行。
- [在线体验](https://paddlejs.baidu.com/ocr)
- [直达教程](https://github.com/PaddlePaddle/Paddle.js/blob/release/v2.2.3/packages/paddlejs-models/ocr/README_cn.md)
- 效果：
<div align="center">
<img src="./paddlejs_demo.gif" width="800">
</div>
\ No newline at end of file
deploy/pdserving/README.md
View file @ aa59fca5
...@@ -36,7 +36,6 @@ PaddleOCR operating environment and Paddle Serving operating environment are nee ...@@ -36,7 +36,6 @@ PaddleOCR operating environment and Paddle Serving operating environment are nee
1. Please prepare PaddleOCR operating environment reference [link](../../doc/doc_ch/installation.md). 1. Please prepare PaddleOCR operating environment reference [link](../../doc/doc_ch/installation.md).
Download the corresponding paddlepaddle whl package according to the environment, it is recommended to install version 2.2.2. Download the corresponding paddlepaddle whl package according to the environment, it is recommended to install version 2.2.2.
2. The steps of PaddleServing operating environment prepare are as follows: 2. The steps of PaddleServing operating environment prepare are as follows:
...@@ -194,6 +193,52 @@ The recognition model is the same. ...@@ -194,6 +193,52 @@ The recognition model is the same.
2021-05-13 03:42:36,979 chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0] 2021-05-13 03:42:36,979 chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
``` ```
## C++ Serving
Service deployment based on python obviously has the advantage of convenient secondary development. However, the real application often needs to pursue better performance. PaddleServing also provides a more performant C++ deployment version.
The C++ service deployment is the same as python in the environment setup and data preparation stages, the difference is when the service is started and the client sends requests.
| Language | Speed ​​| Secondary development | Do you need to compile |
|-----|-----|---------|------------|
| C++ | fast | Slightly difficult | Single model prediction does not need to be compiled, multi-model concatenation needs to be compiled |
| python | general | easy | single-model/multi-model no compilation required |
1. Compile Serving
To improve predictive performance, C++ services also provide multiple model concatenation services. Unlike Python Pipeline services, multiple model concatenation requires the pre - and post-model processing code to be written on the server side, so local recompilation is required to generate serving. Specific may refer to the official document: [how to compile Serving](https://github.com/PaddlePaddle/Serving/blob/v0.8.3/doc/Compile_EN.md)
2. Run the following command to start the service.
```
# Start the service and save the running log in log.txt
python3 -m paddle_serving_server.serve --model ppocrv2_det_serving ppocrv2_rec_serving --op GeneralDetectionOp GeneralInferOp --port 9293 &>log.txt &
```
After the service is successfully started, a log similar to the following will be printed in log.txt
![](./imgs/start_server.png)
3. Send service request
Due to the need for pre and post-processing in the C++Server part, in order to speed up the input to the C++Server is only the base64 encoded string of the picture, it needs to be manually modified
Change the feed_type field and shape field in ppocrv2_det_client/serving_client_conf.prototxt to the following:
```
feed_var {
name: "x"
alias_name: "x"
is_lod_tensor: false
feed_type: 20
shape: 1
}
```
start the client:
```
python3 ocr_cpp_client.py ppocrv2_det_client ppocrv2_rec_client
```
After successfully running, the predicted result of the model will be printed in the cmd window. An example of the result is:
![](./imgs/results.png)
## WINDOWS Users ## WINDOWS Users
Windows does not support Pipeline Serving, if we want to lauch paddle serving on Windows, we should use Web Service, for more infomation please refer to [Paddle Serving for Windows Users](https://github.com/PaddlePaddle/Serving/blob/develop/doc/Windows_Tutorial_EN.md) Windows does not support Pipeline Serving, if we want to lauch paddle serving on Windows, we should use Web Service, for more infomation please refer to [Paddle Serving for Windows Users](https://github.com/PaddlePaddle/Serving/blob/develop/doc/Windows_Tutorial_EN.md)
......
deploy/pdserving/README_CN.md
View file @ aa59fca5
...@@ -6,6 +6,7 @@ PaddleOCR提供2种服务部署方式： ...@@ -6,6 +6,7 @@ PaddleOCR提供2种服务部署方式：
- 基于PaddleHub Serving的部署：代码路径为"`./deploy/hubserving`"，使用方法参考[文档](../../deploy/hubserving/readme.md)； - 基于PaddleHub Serving的部署：代码路径为"`./deploy/hubserving`"，使用方法参考[文档](../../deploy/hubserving/readme.md)；
- 基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，按照本教程使用。 - 基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，按照本教程使用。
# 基于PaddleServing的服务部署 # 基于PaddleServing的服务部署
本文档将介绍如何使用[PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)工具部署PP-OCR动态图模型的pipeline在线服务。 本文档将介绍如何使用[PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)工具部署PP-OCR动态图模型的pipeline在线服务。
...@@ -17,6 +18,8 @@ PaddleOCR提供2种服务部署方式： ...@@ -17,6 +18,8 @@ PaddleOCR提供2种服务部署方式：
更多有关PaddleServing服务化部署框架介绍和使用教程参考[文档](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)。 更多有关PaddleServing服务化部署框架介绍和使用教程参考[文档](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)。
AIStudio演示案例可参考 [基于PaddleServing的OCR服务化部署实战](https://aistudio.baidu.com/aistudio/projectdetail/3630726)。
## 目录 ## 目录
- [环境准备](#环境准备) - [环境准备](#环境准备)
- [模型转换](#模型转换) - [模型转换](#模型转换)
...@@ -30,7 +33,6 @@ PaddleOCR提供2种服务部署方式： ...@@ -30,7 +33,6 @@ PaddleOCR提供2种服务部署方式：
需要准备PaddleOCR的运行环境和Paddle Serving的运行环境。 需要准备PaddleOCR的运行环境和Paddle Serving的运行环境。
- 准备PaddleOCR的运行环境[链接](../../doc/doc_ch/installation.md) - 准备PaddleOCR的运行环境[链接](../../doc/doc_ch/installation.md)
根据环境下载对应的paddlepaddle whl包，推荐安装2.2.2版本
- 准备PaddleServing的运行环境，步骤如下 - 准备PaddleServing的运行环境，步骤如下
...@@ -106,7 +108,7 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv2_rec_infer/ \ ...@@ -106,7 +108,7 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv2_rec_infer/ \
1. 下载PaddleOCR代码，若已下载可跳过此步骤 1. 下载PaddleOCR代码，若已下载可跳过此步骤
``` ```
git clone https://github.com/PaddlePaddle/PaddleOCR git clone https://github.com/PaddlePaddle/PaddleOCR
# 进入到工作目录 # 进入到工作目录
cd PaddleOCR/deploy/pdserving/ cd PaddleOCR/deploy/pdserving/
``` ```
...@@ -132,7 +134,7 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv2_rec_infer/ \ ...@@ -132,7 +134,7 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv2_rec_infer/ \
python3 pipeline_http_client.py python3 pipeline_http_client.py
``` ```
成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为： 成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为：
![](./imgs/results.png) ![](./imgs/pipeline_result.png)
调整 config.yml 中的并发个数获得最大的QPS, 一般检测和识别的并发数为2：1 调整 config.yml 中的并发个数获得最大的QPS, 一般检测和识别的并发数为2：1
``` ```
...@@ -187,6 +189,73 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv2_rec_infer/ \ ...@@ -187,6 +189,73 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv2_rec_infer/ \
2021-05-13 03:42:36,979 chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0] 2021-05-13 03:42:36,979 chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
``` ```
<a name="C++"></a>
## Paddle Serving C++ 部署
基于python的服务部署，显然具有二次开发便捷的优势，然而真正落地应用，往往需要追求更优的性能。PaddleServing 也提供了性能更优的C++部署版本。
C++ 服务部署在环境搭建和数据准备阶段与 python 相同，区别在于启动服务和客户端发送请求时不同。
| 语言 | 速度 | 二次开发 | 是否需要编译 |
|-----|-----|---------|------------|
| C++ | 很快 | 略有难度 | 单模型预测无需编译，多模型串联需要编译 |
| python | 一般 ｜ 容易 | 单模型/多模型 均无需编译|
1. 准备 Serving 环境
为了提高预测性能，C++ 服务同样提供了多模型串联服务。与python pipeline服务不同，多模型串联的过程中需要将模型前后处理代码写在服务端，因此需要在本地重新编译生成serving。
首先需要下载Serving代码库, 把OCR文本检测预处理相关代码替换到Serving库中
```
git clone https://github.com/PaddlePaddle/Serving
cp -rf general_detection_op.cpp Serving/core/general-server/op
```
具体可参考官方文档：[如何编译Serving](https://github.com/PaddlePaddle/Serving/blob/v0.8.3/doc/Compile_CN.md)，注意需要开启 WITH_OPENCV 选项。
完成编译后，注意要安装编译出的三个whl包，并设置SERVING_BIN环境变量。
2. 启动服务可运行如下命令：
一个服务启动两个模型串联，只需要在--model后依次按顺序传入模型文件夹的相对路径，且需要在--op后依次传入自定义C++OP类名称：
```
# 启动服务，运行日志保存在log.txt
python3 -m paddle_serving_server.serve --model ppocrv2_det_serving ppocrv2_rec_serving --op GeneralDetectionOp GeneralInferOp --port 9293 &>log.txt &
```
成功启动服务后，log.txt中会打印类似如下日志
![](./imgs/start_server.png)
3. 发送服务请求：
由于需要在C++Server部分进行前后处理，为了加速传入C++Server的仅仅是图片的base64编码的字符串，故需要手动修改
ppocrv2_det_client/serving_client_conf.prototxt 中 feed_type 字段 和 shape 字段，修改成如下内容：
```
feed_var {
name: "x"
alias_name: "x"
is_lod_tensor: false
feed_type: 20
shape: 1
}
```
启动客户端
```
python3 ocr_cpp_client.py ppocrv2_det_client ppocrv2_rec_client
```
成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为：
![](./imgs/results.png)
在浏览器中输入服务器 ip:端口号，可以看到当前服务的实时QPS。(端口号范围需要是8000-9000)
在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 峰值可达到51左右,约为pipeline的 2.12 倍。
![](./imgs/c++_qps.png)
<a name="Windows用户"></a> <a name="Windows用户"></a>
## Windows用户 ## Windows用户
......
deploy/pdserving/general_detection_op.cpp 0 → 100644
View file @ aa59fca5
This diff is collapsed.
deploy/pdserving/ocr_cpp_client.py
View file @ aa59fca5
...@@ -45,10 +45,8 @@ for img_file in os.listdir(test_img_dir): ...@@ -45,10 +45,8 @@ for img_file in os.listdir(test_img_dir):
image_data = file.read() image_data = file.read()
image = cv2_to_base64(image_data) image = cv2_to_base64(image_data)
res_list = [] res_list = []
#print(image)
fetch_map = client.predict( fetch_map = client.predict(
feed={"x": image}, fetch=["save_infer_model/scale_0.tmp_1"], batch=True) feed={"x": image}, fetch=["save_infer_model/scale_0.tmp_1"], batch=True)
print("fetrch map:", fetch_map)
one_batch_res = ocr_reader.postprocess(fetch_map, with_score=True) one_batch_res = ocr_reader.postprocess(fetch_map, with_score=True)
for res in one_batch_res: for res in one_batch_res:
res_list.append(res[0]) res_list.append(res[0])
......
deploy/pdserving/pipeline_http_client.py
View file @ aa59fca5
This diff is collapsed.
deploy/pdserving/web_service.py
View file @ aa59fca5
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment