init

2024-01-17 15:18:16 +08:00
commit 1bd4060e68
1514 changed files with 193581 additions and 0 deletions
@@ -0,0 +1,102 @@
+[English](README.md) | 简体中文
+
+# PaddleOCR 模型在SOPHGO上部署方案-FastDeploy
+
+## 1. 说明  
+PaddleOCR支持通过FastDeploy在SOPHGO上部署相关模型.
+
+## 2.支持模型列表
+
+下表中的模型下载链接由PaddleOCR模型库提供, 详见[PP-OCR系列模型列表](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/doc/doc_ch/models_list.md)
+
+| PaddleOCR版本 | 文本框检测 | 方向分类模型 | 文字识别 |字典文件| 说明 |
+|:----|:----|:----|:----|:----|:--------|
+| ch_PP-OCRv3[推荐] |[ch_PP-OCRv3_det](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar) | [ch_ppocr_mobile_v2.0_cls](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) | [ch_PP-OCRv3_rec](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar) | [ppocr_keys_v1.txt](https://bj.bcebos.com/paddlehub/fastdeploy/ppocr_keys_v1.txt) | OCRv3系列原始超轻量模型，支持中英文、多语种文本检测 |
+| en_PP-OCRv3[推荐] |[en_PP-OCRv3_det](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) | [ch_ppocr_mobile_v2.0_cls](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) | [en_PP-OCRv3_rec](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar) | [en_dict.txt](https://bj.bcebos.com/paddlehub/fastdeploy/en_dict.txt) | OCRv3系列原始超轻量模型，支持英文与数字识别，除检测模型和识别模型的训练数据与中文模型不同以外，无其他区别 |
+| ch_PP-OCRv2 |[ch_PP-OCRv2_det](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar) | [ch_ppocr_mobile_v2.0_cls](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) | [ch_PP-OCRv2_rec](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) | [ppocr_keys_v1.txt](https://bj.bcebos.com/paddlehub/fastdeploy/ppocr_keys_v1.txt) | OCRv2系列原始超轻量模型，支持中英文、多语种文本检测 |
+| ch_PP-OCRv2_mobile |[ch_ppocr_mobile_v2.0_det](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) | [ch_ppocr_mobile_v2.0_cls](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) | [ch_ppocr_mobile_v2.0_rec](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) | [ppocr_keys_v1.txt](https://bj.bcebos.com/paddlehub/fastdeploy/ppocr_keys_v1.txt) | OCRv2系列原始超轻量模型，支持中英文、多语种文本检测,比PPOCRv2更加轻量 |
+| ch_PP-OCRv2_server |[ch_ppocr_server_v2.0_det](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) | [ch_ppocr_mobile_v2.0_cls](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) | [ch_ppocr_server_v2.0_rec](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) |[ppocr_keys_v1.txt](https://bj.bcebos.com/paddlehub/fastdeploy/ppocr_keys_v1.txt) | OCRv2服务器系列模型, 支持中英文、多语种文本检测，比超轻量模型更大，但效果更好|
+
+## 3. 准备PP-OCR推理模型以及转换模型
+
+PP-OCRv3包括文本检测模型（ch_PP-OCRv3_det）、方向分类模型（ch_ppocr_mobile_v2.0_cls）、文字识别模型（ch_PP-OCRv3_rec）  
+SOPHGO-TPU部署模型前需要将以上Paddle模型转换成bmodel模型，我们以ch_PP-OCRv3_det模型为例，具体步骤如下:
+- 下载Paddle模型[ch_PP-OCRv3_det](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar)
+- Pddle模型转换为ONNX模型，请参考[Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX)
+- ONNX模型转换bmodel模型的过程，请参考[TPU-MLIR](https://github.com/sophgo/tpu-mlir)
+下面我们提供一个example, 供用户参考，完成模型的转换.
+
+### 3.1 下载ch_PP-OCRv3_det模型,并转换为ONNX模型
+```shell
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar
+tar xvf ch_PP-OCRv3_det_infer.tar
+
+# 修改ch_PP-OCRv3_det模型的输入shape，由动态输入变成固定输入
+python paddle_infer_shape.py --model_dir ch_PP-OCRv3_det_infer \
+                             --model_filename inference.pdmodel \
+                             --params_filename inference.pdiparams \
+                             --save_dir ch_PP-OCRv3_det_infer_fix \
+                             --input_shape_dict="{'x':[1,3,960,608]}"
+
+# 请用户自行安装最新发布版本的paddle2onnx, 转换模型到ONNX格式的模型
+paddle2onnx --model_dir ch_PP-OCRv3_det_infer_fix \
+            --model_filename inference.pdmodel \
+            --params_filename inference.pdiparams \
+            --save_file ch_PP-OCRv3_det_infer_fix.onnx \
+            --enable_dev_version True
+```
+
+### 3.2 导出bmodel模型
+
+以转换BM1684x的bmodel模型为例子，我们需要下载[TPU-MLIR](https://github.com/sophgo/tpu-mlir)工程，安装过程具体参见[TPU-MLIR文档](https://github.com/sophgo/tpu-mlir/blob/master/README.md)。
+#### 3.2.1    安装
+``` shell
+docker pull sophgo/tpuc_dev:latest
+
+# myname1234是一个示例，也可以设置其他名字
+docker run --privileged --name myname1234 -v $PWD:/workspace -it sophgo/tpuc_dev:latest
+
+source ./envsetup.sh
+./build.sh
+```
+
+#### 3.2.2    ONNX模型转换为bmodel模型
+``` shell
+mkdir ch_PP-OCRv3_det && cd ch_PP-OCRv3_det
+
+#在该文件中放入测试图片，同时将上一步转换的ch_PP-OCRv3_det_infer_fix.onnx放入该文件夹中
+cp -rf ${REGRESSION_PATH}/dataset/COCO2017 .
+cp -rf ${REGRESSION_PATH}/image .
+#放入onnx模型文件ch_PP-OCRv3_det_infer_fix.onnx
+
+mkdir workspace && cd workspace
+
+#将ONNX模型转换为mlir模型，其中参数--output_names可以通过NETRON查看
+model_transform.py \
+    --model_name ch_PP-OCRv3_det \
+    --model_def ../ch_PP-OCRv3_det_infer_fix.onnx \
+    --input_shapes [[1,3,960,608]] \
+    --mean 0.0,0.0,0.0 \
+    --scale 0.0039216,0.0039216,0.0039216 \
+    --keep_aspect_ratio \
+    --pixel_format rgb \
+    --output_names sigmoid_0.tmp_0 \
+    --test_input ../image/dog.jpg \
+    --test_result ch_PP-OCRv3_det_top_outputs.npz \
+    --mlir ch_PP-OCRv3_det.mlir
+
+#将mlir模型转换为BM1684x的F32 bmodel模型
+model_deploy.py \
+  --mlir ch_PP-OCRv3_det.mlir \
+  --quantize F32 \
+  --chip bm1684x \
+  --test_input ch_PP-OCRv3_det_in_f32.npz \
+  --test_reference ch_PP-OCRv3_det_top_outputs.npz \
+  --model ch_PP-OCRv3_det_1684x_f32.bmodel
+```
+最终获得可以在BM1684x上能够运行的bmodel模型ch_PP-OCRv3_det_1684x_f32.bmodel。按照上面同样的方法，可以将ch_ppocr_mobile_v2.0_cls，ch_PP-OCRv3_rec转换为bmodel的格式。如果需要进一步对模型进行加速，可以将ONNX模型转换为INT8 bmodel，具体步骤参见[TPU-MLIR文档](https://github.com/sophgo/tpu-mlir/blob/master/README.md)。
+
+
+## 4. 详细部署的部署示例  
+- [Python部署](python)
+- [C++部署](cpp)
@@ -0,0 +1,13 @@
+PROJECT(infer_demo C CXX)
+CMAKE_MINIMUM_REQUIRED (VERSION 3.10)
+# 指定下载解压后的fastdeploy库路径
+option(FASTDEPLOY_INSTALL_DIR "Path of downloaded fastdeploy sdk.")
+
+include(${FASTDEPLOY_INSTALL_DIR}/FastDeploy.cmake)
+
+# 添加FastDeploy依赖头文件
+include_directories(${FASTDEPLOY_INCS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
@@ -0,0 +1,66 @@
+[English](README_CN.md) | 简体中文
+# PP-OCRv3 SOPHGO C++部署示例
+本目录下提供`infer.cc`快速完成PPOCRv3模型在SOPHGO BM1684x板子上加速部署的示例。
+
+## 1. 部署环境准备
+在部署前，需自行编译基于SOPHGO硬件的预测库，参考文档[SOPHGO硬件部署环境](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install#算能硬件部署环境)
+
+## 2. 生成基本目录文件
+
+该例程由以下几个部分组成
+```text
+.
+├── CMakeLists.txt
+├── fastdeploy-sophgo  # 编译好的SDK文件夹
+├── image  # 存放图片的文件夹
+├── infer.cc
+└── model  # 存放模型文件的文件夹
+```
+
+## 3.部署示例
+
+### 3.1 下载部署示例代码
+```bash
+# 下载部署示例代码
+git clone https://github.com/PaddlePaddle/FastDeploy.git
+cd  FastDeploy/examples/vision/ocr/PP-OCR/sophgo/cpp
+
+# 如果您希望从PaddleOCR下载示例代码，请运行
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+# 注意：如果当前分支找不到下面的fastdeploy测试代码，请切换到dygraph分支
+git checkout dygraph
+cd PaddleOCR/deploy/fastdeploy/sophgo/cpp
+```
+
+### 3.2 拷贝bmodel模型文至model文件夹
+将Paddle模型转换为SOPHGO bmodel模型，转换步骤参考[文档](../README.md). 将转换后的SOPHGO bmodel模型文件拷贝至model中.
+
+### 3.3 准备测试图片至image文件夹，以及字典文件
+```bash
+wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/doc/imgs/12.jpg
+cp 12.jpg image/
+
+wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_keys_v1.txt
+```
+
+### 3.4 编译example
+
+```bash
+cd build
+cmake .. -DFASTDEPLOY_INSTALL_DIR=${PWD}/fastdeploy-0.0.3
+make
+```
+
+### 3.5 运行例程
+
+```bash
+./infer_demo model ./ppocr_keys_v1.txt image/12.jpeg
+```
+
+
+## 4. 更多指南
+
+- [PP-OCR系列 C++ API查阅](https://www.paddlepaddle.org.cn/fastdeploy-api-doc/cpp/html/namespacefastdeploy_1_1vision_1_1ocr.html)
+- [FastDeploy部署PaddleOCR模型概览](../../)
+- [PP-OCRv3 Python部署](../python)
+- 如果用户想要调整前后处理超参数、单独使用文字检测识别模型、使用其他模型等，更多详细文档与说明请参考[PP-OCR系列在CPU/GPU上的部署](../../cpu-gpu/cpp/README.md)
@@ -0,0 +1,136 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision.h"
+#ifdef WIN32
+const char sep = '\\';
+#else
+const char sep = '/';
+#endif
+
+void InitAndInfer(const std::string &det_model_dir,
+                  const std::string &rec_label_file,
+                  const std::string &image_file,
+                  const fastdeploy::RuntimeOption &option) {
+  auto det_model_file =
+      det_model_dir + sep + "ch_PP-OCRv3_det_1684x_f32.bmodel";
+  auto det_params_file = det_model_dir + sep + "";
+
+  auto cls_model_file =
+      det_model_dir + sep + "ch_ppocr_mobile_v2.0_cls_1684x_f32.bmodel";
+  auto cls_params_file = det_model_dir + sep + "";
+
+  auto rec_model_file =
+      det_model_dir + sep + "ch_PP-OCRv3_rec_1684x_f32.bmodel";
+  auto rec_params_file = det_model_dir + sep + "";
+
+  auto format = fastdeploy::ModelFormat::SOPHGO;
+
+  auto det_option = option;
+  auto cls_option = option;
+  auto rec_option = option;
+
+  // The cls and rec model can inference a batch of images now.
+  // User could initialize the inference batch size and set them after create
+  // PPOCR model.
+  int cls_batch_size = 1;
+  int rec_batch_size = 1;
+
+  // If use TRT backend, the dynamic shape will be set as follow.
+  // We recommend that users set the length and height of the detection model to
+  // a multiple of 32. We also recommend that users set the Trt input shape as
+  // follow.
+  det_option.SetTrtInputShape("x", {1, 3, 64, 64}, {1, 3, 640, 640},
+                              {1, 3, 960, 960});
+  cls_option.SetTrtInputShape("x", {1, 3, 48, 10}, {cls_batch_size, 3, 48, 320},
+                              {cls_batch_size, 3, 48, 1024});
+  rec_option.SetTrtInputShape("x", {1, 3, 48, 10}, {rec_batch_size, 3, 48, 320},
+                              {rec_batch_size, 3, 48, 2304});
+
+  // Users could save TRT cache file to disk as follow.
+  // det_option.SetTrtCacheFile(det_model_dir + sep + "det_trt_cache.trt");
+  // cls_option.SetTrtCacheFile(cls_model_dir + sep + "cls_trt_cache.trt");
+  // rec_option.SetTrtCacheFile(rec_model_dir + sep + "rec_trt_cache.trt");
+
+  auto det_model = fastdeploy::vision::ocr::DBDetector(
+      det_model_file, det_params_file, det_option, format);
+  auto cls_model = fastdeploy::vision::ocr::Classifier(
+      cls_model_file, cls_params_file, cls_option, format);
+  auto rec_model = fastdeploy::vision::ocr::Recognizer(
+      rec_model_file, rec_params_file, rec_label_file, rec_option, format);
+
+  // Users could enable static shape infer for rec model when deploy PP-OCR on
+  // hardware which can not support dynamic shape infer well, like Huawei Ascend
+  // series.
+  rec_model.GetPreprocessor().SetStaticShapeInfer(true);
+  rec_model.GetPreprocessor().SetRecImageShape({3, 48, 584});
+
+  assert(det_model.Initialized());
+  assert(cls_model.Initialized());
+  assert(rec_model.Initialized());
+
+  // The classification model is optional, so the PP-OCR can also be connected
+  // in series as follows auto ppocr_v3 =
+  // fastdeploy::pipeline::PPOCRv3(&det_model, &rec_model);
+  auto ppocr_v3 =
+      fastdeploy::pipeline::PPOCRv3(&det_model, &cls_model, &rec_model);
+
+  // Set inference batch size for cls model and rec model, the value could be -1
+  // and 1 to positive infinity. When inference batch size is set to -1, it
+  // means that the inference batch size of the cls and rec models will be the
+  // same as the number of boxes detected by the det model.
+  ppocr_v3.SetClsBatchSize(cls_batch_size);
+  ppocr_v3.SetRecBatchSize(rec_batch_size);
+
+  if (!ppocr_v3.Initialized()) {
+    std::cerr << "Failed to initialize PP-OCR." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  auto im_bak = im.clone();
+
+  fastdeploy::vision::OCRResult result;
+  if (!ppocr_v3.Predict(&im, &result)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+
+  std::cout << result.Str() << std::endl;
+
+  auto vis_im = fastdeploy::vision::VisOcr(im_bak, result);
+  cv::imwrite("vis_result.jpg", vis_im);
+  std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
+}
+
+int main(int argc, char *argv[]) {
+  if (argc < 4) {
+    std::cout << "Usage: infer_demo path/to/model "
+                 "path/to/rec_label_file path/to/image "
+                 "e.g ./infer_demo ./ocr_bmodel "
+                 "./ppocr_keys_v1.txt ./12.jpg"
+              << std::endl;
+    return -1;
+  }
+
+  fastdeploy::RuntimeOption option;
+  option.UseSophgo();
+  option.UseSophgoBackend();
+
+  std::string model_dir = argv[1];
+  std::string rec_label_file = argv[2];
+  std::string test_image = argv[3];
+  InitAndInfer(model_dir, rec_label_file, test_image, option);
+  return 0;
+}
@@ -0,0 +1,52 @@
+[English](README.md) | 简体中文
+# PP-OCRv3 SOPHGO Python部署示例
+本目录下提供`infer.py`快速完成 PP-OCRv3 在SOPHGO TPU上部署的示例。
+
+## 1. 部署环境准备
+
+在部署前，需自行编译基于算能硬件的FastDeploy python wheel包并安装，参考文档[算能硬件部署环境](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install#算能硬件部署环境)
+
+
+## 2.运行部署示例
+
+### 2.1 模型准备
+将Paddle模型转换为SOPHGO bmodel模型, 转换步骤参考[文档](../README.md)  
+
+### 2.2 开始部署
+```bash
+# 下载部署示例代码
+git clone https://github.com/PaddlePaddle/FastDeploy.git
+cd  FastDeploy/examples/vision/ocr/PP-OCR/sophgo/python
+
+# 如果您希望从PaddleOCR下载示例代码，请运行
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+# 注意：如果当前分支找不到下面的fastdeploy测试代码，请切换到dygraph分支
+git checkout dygraph
+cd PaddleOCR/deploy/fastdeploy/sophgo/python
+
+# 下载图片
+wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/doc/imgs/12.jpg
+
+#下载字典文件
+wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_keys_v1.txt
+
+# 推理
+python3 infer.py --det_model ocr_bmodel/ch_PP-OCRv3_det_1684x_f32.bmodel \
+                 --cls_model ocr_bmodel/ch_ppocr_mobile_v2.0_cls_1684x_f32.bmodel \
+                 --rec_model ocr_bmodel/ch_PP-OCRv3_rec_1684x_f32.bmodel \
+                 --rec_label_file ../ppocr_keys_v1.txt \  
+                 --image ../12.jpg
+
+# 运行完成后返回结果如下所示
+det boxes: [[42,413],[483,391],[484,428],[43,450]]rec text: 上海斯格威铂尔大酒店 rec score:0.952958 cls label: 0 cls score: 1.000000
+det boxes: [[187,456],[399,448],[400,480],[188,488]]rec text: 打浦路15号 rec score:0.897335 cls label: 0 cls score: 1.000000
+det boxes: [[23,507],[513,488],[515,529],[24,548]]rec text: 绿洲仕格维花园公寓 rec score:0.994589 cls label: 0 cls score: 1.000000
+det boxes: [[74,553],[427,542],[428,571],[75,582]]rec text: 打浦路252935号 rec score:0.900663 cls label: 0 cls score: 1.000000
+
+可视化结果保存在sophgo_result.jpg中
+```
+
+## 3. 其它文档
+- [PP-OCRv3 C++部署](../cpp)
+- [转换 PP-OCRv3 SOPHGO模型文档](../README.md)
+- 如果用户想要调整前后处理超参数、单独使用文字检测识别模型、使用其他模型等，更多详细文档与说明请参考[PP-OCR系列在CPU/GPU上的部署](../../cpu-gpu/cpp/README.md)
@@ -0,0 +1,116 @@
+import fastdeploy as fd
+import cv2
+import os
+
+
+def parse_arguments():
+    import argparse
+    import ast
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--det_model", required=True, help="Path of Detection model of PPOCR.")
+    parser.add_argument(
+        "--cls_model",
+        required=True,
+        help="Path of Classification model of PPOCR.")
+    parser.add_argument(
+        "--rec_model",
+        required=True,
+        help="Path of Recognization model of PPOCR.")
+    parser.add_argument(
+        "--rec_label_file",
+        required=True,
+        help="Path of Recognization label of PPOCR.")
+    parser.add_argument(
+        "--image", type=str, required=True, help="Path of test image file.")
+
+    return parser.parse_args()
+
+
+args = parse_arguments()
+
+# 配置runtime，加载模型
+runtime_option = fd.RuntimeOption()
+runtime_option.use_sophgo()
+
+# Detection模型, 检测文字框
+det_model_file = args.det_model
+det_params_file = ""
+# Classification模型，方向分类，可选
+cls_model_file = args.cls_model
+cls_params_file = ""
+# Recognition模型，文字识别模型
+rec_model_file = args.rec_model
+rec_params_file = ""
+rec_label_file = args.rec_label_file
+
+# PPOCR的cls和rec模型现在已经支持推理一个Batch的数据
+# 定义下面两个变量后, 可用于设置trt输入shape, 并在PPOCR模型初始化后, 完成Batch推理设置
+cls_batch_size = 1
+rec_batch_size = 1
+
+# 当使用TRT时，分别给三个模型的runtime设置动态shape,并完成模型的创建.
+# 注意: 需要在检测模型创建完成后，再设置分类模型的动态输入并创建分类模型, 识别模型同理.
+# 如果用户想要自己改动检测模型的输入shape, 我们建议用户把检测模型的长和高设置为32的倍数.
+det_option = runtime_option
+det_option.set_trt_input_shape("x", [1, 3, 64, 64], [1, 3, 640, 640],
+                               [1, 3, 960, 960])
+# 用户可以把TRT引擎文件保存至本地
+# det_option.set_trt_cache_file(args.det_model  + "/det_trt_cache.trt")
+det_model = fd.vision.ocr.DBDetector(
+    det_model_file,
+    det_params_file,
+    runtime_option=det_option,
+    model_format=fd.ModelFormat.SOPHGO)
+
+cls_option = runtime_option
+cls_option.set_trt_input_shape("x", [1, 3, 48, 10],
+                               [cls_batch_size, 3, 48, 320],
+                               [cls_batch_size, 3, 48, 1024])
+# 用户可以把TRT引擎文件保存至本地
+# cls_option.set_trt_cache_file(args.cls_model  + "/cls_trt_cache.trt")
+cls_model = fd.vision.ocr.Classifier(
+    cls_model_file,
+    cls_params_file,
+    runtime_option=cls_option,
+    model_format=fd.ModelFormat.SOPHGO)
+
+rec_option = runtime_option
+rec_option.set_trt_input_shape("x", [1, 3, 48, 10],
+                               [rec_batch_size, 3, 48, 320],
+                               [rec_batch_size, 3, 48, 2304])
+# 用户可以把TRT引擎文件保存至本地
+# rec_option.set_trt_cache_file(args.rec_model  + "/rec_trt_cache.trt")
+rec_model = fd.vision.ocr.Recognizer(
+    rec_model_file,
+    rec_params_file,
+    rec_label_file,
+    runtime_option=rec_option,
+    model_format=fd.ModelFormat.SOPHGO)
+
+# 创建PP-OCR，串联3个模型，其中cls_model可选，如无需求，可设置为None
+ppocr_v3 = fd.vision.ocr.PPOCRv3(
+    det_model=det_model, cls_model=cls_model, rec_model=rec_model)
+
+# 需要使用下行代码, 来启用rec模型的静态shape推理，这里rec模型的静态输入为[3, 48, 584]
+rec_model.preprocessor.static_shape_infer = True
+rec_model.preprocessor.rec_image_shape = [3, 48, 584]
+
+# 给cls和rec模型设置推理时的batch size
+# 此值能为-1, 和1到正无穷
+# 当此值为-1时, cls和rec模型的batch size将默认和det模型检测出的框的数量相同
+ppocr_v3.cls_batch_size = cls_batch_size
+ppocr_v3.rec_batch_size = rec_batch_size
+
+# 预测图片准备
+im = cv2.imread(args.image)
+
+#预测并打印结果
+result = ppocr_v3.predict(im)
+
+print(result)
+
+# 可视化结果
+vis_im = fd.vision.vis_ppocr(im, result)
+cv2.imwrite("sophgo_result.jpg", vis_im)
+print("Visualized result save in ./sophgo_result.jpg")