apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-04 16:22:57 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Other] Add detection, segmentation and OCR examples for Ascend deploy. (#983)

Browse Source 
* Add Huawei Ascend NPU deploy through PaddleLite CANN

* Add NNAdapter interface for paddlelite

* Modify Huawei Ascend Cmake

* Update way for compiling Huawei Ascend NPU deployment

* remove UseLiteBackend in UseCANN

* Support compile python whlee

* Change names of nnadapter API

* Add nnadapter pybind and remove useless API

* Support Python deployment on Huawei Ascend NPU

* Add models suppor for ascend

* Add PPOCR rec reszie for ascend

* fix conflict for ascend

* Rename CANN to Ascend

* Rename CANN to Ascend

* Improve ascend

* fix ascend bug

* improve ascend docs

* improve ascend docs

* improve ascend docs

* Improve Ascend

* Improve Ascend

* Move ascend python demo

* Imporve ascend

* Improve ascend

* Improve ascend

* Improve ascend

* Improve ascend

* Imporve ascend

* Imporve ascend

* Improve ascend

* acc eval script

* acc eval

* remove acc_eval from branch huawei

* Add detection and segmentation examples for Ascend deployment

* Add detection and segmentation examples for Ascend deployment

* Add PPOCR example for ascend deploy

* Imporve paddle lite compiliation

* Add FlyCV doc

* Add FlyCV doc

* Add FlyCV doc

* Imporve Ascend docs

* Imporve Ascend docs
This commit is contained in:
yunyaoXYY
2023-01-04 10:01:23 +08:00
committed by GitHub
parent 2cfd331889
commit 58d63f3e90
57 changed files with 694 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
cmake/ascend.cmake
View File

@@ -16,13 +16,4 @@ else ()
if(NOT PADDLELITE_URL)
set(PADDLELITE_URL "https://bj.bcebos.com/fastdeploy/test/lite-linux_arm64_huawei_ascend_npu_python_1207.tgz")
endif()
execute_process(COMMAND sh -c "ls *.so*" WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/third_libs/install/paddlelite/lib
COMMAND sh -c "xargs ${PATCHELF_EXE} --set-rpath '$ORIGIN'" WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/third_libs/install/paddlelite/lib
RESULT_VARIABLE result
OUTPUT_VARIABLE curr_out
ERROR_VARIABLE curr_out)
if(ret EQUAL "1")
message(FATAL_ERROR "Failed to patchelf Paddle Lite libraries when using Ascend.")
endif()
message(STATUS "result:${result} out:${curr_out}")
endif()

4
cmake/paddlelite.cmake
View File

@@ -114,6 +114,10 @@ else()
BUILD_BYPRODUCTS ${PADDLELITE_LIB})
endif()
if(UNIX AND (NOT APPLE) AND (NOT ANDROID))
add_custom_target(patchelf_paddle_lite ALL COMMAND bash -c "PATCHELF_EXE=${PATCHELF_EXE} python ${PROJECT_SOURCE_DIR}/scripts/patch_paddle_lite.py ${PADDLELITE_INSTALL_DIR}/lib/" DEPENDS ${LIBRARY_NAME})
endif()
add_library(external_paddle_lite STATIC IMPORTED GLOBAL)
set_property(TARGET external_paddle_lite PROPERTY IMPORTED_LOCATION ${PADDLELITE_LIB})
add_dependencies(external_paddle_lite ${PADDLELITE_PROJECT})

10
docs/cn/build_and_install/huawei_ascend.md
View File

@@ -94,6 +94,12 @@ python setup.py bdist_wheel
#编译完成后,请用户自行安装当前目录的dist文件夹内的whl包.
```
## 五.昇腾部署时开启FlyCV
[FlyCV](https://github.com/PaddlePaddle/FlyCV) 是一款高性能计算机图像处理库, 针对ARM架构做了很多优化, 相比其他图像处理库性能更为出色.
FastDeploy现在已经集成FlyCV, 用户可以在支持的硬件平台上使用FlyCV, 实现模型端到端推理性能的加速.
模型端到端推理中, 预处理和后处理阶段为CPU计算, 当用户使用ARM CPU + 昇腾的硬件平台时, 我们推荐用户使用FlyCV, 可以实现端到端的推理性能加速, 详见[FLyCV使用文档](./boost_cv_by_flycv.md).
- 华为昇腾NPU 上使用C++部署 PaddleClas 分类模型请参考：[PaddleClas 华为升腾NPU C++ 部署示例](../../../examples/vision/classification/paddleclas/ascend/cpp/README.md)
- 华为昇腾NPU 上使用Python部署 PaddleClas 分类模型请参考：[PaddleClas 华为升腾NPU Python 部署示例](../../../examples/vision/classification/paddleclas/ascend/python/README.md)
## 六.昇腾部署Demo参考
- 华为昇腾NPU 上使用C++部署 PaddleClas 分类模型请参考：[PaddleClas 华为升腾NPU C++ 部署示例](../../../examples/vision/classification/paddleclas/cpp/README.md)
- 华为昇腾NPU 上使用Python部署 PaddleClas 分类模型请参考：[PaddleClas 华为升腾NPU Python 部署示例](../../../examples/vision/classification/paddleclas/python/README.md)

68
docs/cn/faq/boost_cv_by_flycv.md Normal file
View File

@@ -0,0 +1,68 @@
[English](../../en/faq/boost_cv_by_flycv.md) | 中文
# 使用FlyCV加速端到端推理性能
[FlyCV](https://github.com/PaddlePaddle/FlyCV) 是一款高性能计算机图像处理库, 针对ARM架构做了很多优化, 相比其他图像处理库性能更为出色.
FastDeploy现在已经集成FlyCV, 用户可以在支持的硬件平台上使用FlyCV, 实现模型端到端推理性能的加速.
## 已支持的系统与硬件架构
| 系统 | 硬件架构 |
| :-----------| :-------- |
| Android | armeabi-v7a, arm64-v8a |
| Linux | aarch64, armhf, x86_64|
## 使用方式
使用FlyCV,首先需要在编译时开启FlyCV编译选项,之后在部署时新增一行代码即可开启.
本文以Linux系统为例,说明如何开启FlyCV编译选项, 之后在部署时, 新增一行代码使用FlyCV.
用户可以按照如下方式,在编译预测库时,开启FlyCV编译选项.
```bash
# 编译C++预测库时, 开启FlyCV编译选项.
-DENABLE_VISION=ON \
# 在编译Python预测库时, 开启FlyCV编译选项
export ENABLE_FLYCV=ON
```
用户可以按照如下方式,在部署代码中新增一行代码启用FlyCV.
```bash
# C++部署代码.
# 新增一行代码启用FlyCV
fastdeploy::vision::EnableFlyCV();
# 其他部署代码...(以昇腾部署为例)
fastdeploy::RuntimeOption option;
option.UseAscend();
...
# Python部署代码
# 新增一行代码启用FlyCV
fastdeploy.vision.enable_flycv()
# 其他部署代码...(以昇腾部署为例)
runtime_option = build_option()
option.use_ascend()
...
```
## 部分平台FlyCV 端到端性能数据
鲲鹏920 CPU + Atlas 300I Pro 推理卡.
| 模型 | OpenCV 端到端性能(ms) | FlyCV 端到端性能(ms) |
| :-----------| :-------- | :-------- |
| ResNet50 | 2.78 | 1.63 |
| PP-LCNetV2 | 2.50 | 1.39 |
| YOLOv7 | 27.00 | 21.36 |
| PP_HumanSegV2_Lite | 2.76 | 2.10 |
瑞芯微RV1126.
| 模型 | OpenCV 端到端性能(ms) | FlyCV 端到端性能(ms) |
| :-----------| :-------- | :-------- |
| ResNet50 | 9.23 | 6.01 |
| mobilenetv1_ssld_量化模型 | 9.23 | 6.01 |
| yolov5s_量化模型 | 28.33 | 14.25 |
| PP_LiteSeg_量化模型 | 132.25 | 60.31 |

2
docs/cn/faq/use_sdk_on_ascend.md
View File

@@ -1,4 +1,4 @@
[English](../../en/faq/use_sdk_on_linux.md) | 中文
[English](../../en/faq/use_sdk_on_ascend.md) | 中文
# Linux上使用C++在华为昇腾部署

12
docs/en/build_and_install/huawei_ascend.md
View File

@@ -93,6 +93,14 @@ python setup.py bdist_wheel
#After the compilation is complete, please install the whl package in the dist folder of the current directory.
```
Deploying PaddleClas Classification Model on Huawei Ascend NPU using C++ please refer to: [PaddleClas Huawei Ascend NPU C++ Deployment Example](../../../examples/vision/classification/paddleclas/ascend/cpp/README.md)
## Enable FlyCV for Ascend deployment
Deploying PaddleClas classification model on Huawei Ascend NPU using Python please refer to: [PaddleClas Huawei Ascend NPU Python Deployment Example](../../../examples/vision/classification/paddleclas/ascend/python/README.md)
[FlyCV](https://github.com/PaddlePaddle/FlyCV) is a high performance computer image processing library, providing better performance than other image processing libraries, especially in the ARM architecture.
FastDeploy is now integrated with FlyCV, allowing users to use FlyCV on supported hardware platforms to accelerate model end-to-end inference performance.
In end-to-end model inference, the pre-processing and post-processing phases are CPU computation, we recommend using FlyCV for end-to-end inference performance acceleration when you are using ARM CPU + Ascend hardware platform. See [Enable FlyCV](./boost_cv_by_flycv.md) documentation for details.
## Deployment demo reference
- Deploying PaddleClas Classification Model on Huawei Ascend NPU using C++ please refer to: [PaddleClas Huawei Ascend NPU C++ Deployment Example](../../../examples/vision/classification/paddleclas/cpp/README.md)
- Deploying PaddleClas classification model on Huawei Ascend NPU using Python please refer to: [PaddleClas Huawei Ascend NPU Python Deployment Example](../../../examples/vision/classification/paddleclas/python/README.md)

66
docs/en/faq/boost_cv_by_flycv.md Normal file
View File

@@ -0,0 +1,66 @@
[简体中文](../../cn/faq/boost_cv_by_flycv.md) | English
# Accelerate end-to-end inference performance using FlyCV
[FlyCV](https://github.com/PaddlePaddle/FlyCV) is a high performance computer image processing library, providing better performance than other image processing libraries, especially in the ARM architecture.
FastDeploy is now integrated with FlyCV, allowing users to use FlyCV on supported hardware platforms to accelerate model end-to-end inference performance.
## Supported OS and Architectures
| OS | Architectures |
| :-----------| :-------- |
| Android | armeabi-v7a, arm64-v8a |
| Linux | aarch64, armhf, x86_64|
## Usage
To use FlyCV, you first need to turn on the FlyCV compile option at compile time, and then add a new line of code to turn it on.
This article uses Linux as an example to show how to enable the FlyCV compile option, and then add a new line of code to use FlyCV during deployment.
You can turn on the FlyCV compile option when compiling the FastDeploy library as follows.
```bash
# When compiling C++ libraries
-DENABLE_VISION=ON
# When compiling Python libraries
export ENABLE_FLYCV=ON
```
You can enable FlyCV by adding a new line of code to the deployment code as follows.
```bash
# C++ code
fastdeploy::vision::EnableFlyCV();
# Other..(e.g. With Huawei Ascend)
fastdeploy::RuntimeOption option;
option.UseAscend();
...
# Python code
fastdeploy.vision.enable_flycv()
# Other..(e.g. With Huawei Ascend)
runtime_option = build_option()
option.use_ascend()
...
```
## Some Platforms FlyCV End-to-End Inference Performance
KunPeng 920 CPU + Atlas 300I Pro.
| Model | OpenCV E2E Performance(ms) | FlyCV E2E Performance(ms) |
| :-----------| :-------- | :-------- |
| ResNet50 | 2.78 | 1.63 |
| PP-LCNetV2 | 2.50 | 1.39 |
| YOLOv7 | 27.00 | 21.36 |
| PP_HumanSegV2_Lite | 2.76 | 2.10 |
Rockchip RV1126.
| Model | OpenCV E2E Performance(ms) | FlyCV E2E Performance(ms) |
| :-----------| :-------- | :-------- |
| ResNet50 | 9.23 | 6.01 |
| mobilenetv1_ssld_量化模型 | 9.23 | 6.01 |
| yolov5s_量化模型 | 28.33 | 14.25 |
| PP_LiteSeg_量化模型 | 132.25 | 60.31 |

2
docs/en/faq/use_sdk_on_ascend.md
View File

@@ -1,4 +1,4 @@
[简体中文](../../cn/faq/use_sdk_on_linux.md) | English
[简体中文](../../cn/faq/use_sdk_on_ascend.md) | English
# # Linux deployment with C++ on Huawei Ascend

2
examples/vision/detection/paddledetection/cpp/README.md
View File

@@ -34,6 +34,8 @@ tar xvf ppyoloe_crn_l_300e_coco.tgz
./infer_ppyoloe_demo ./ppyoloe_crn_l_300e_coco 000000014439.jpg 2
# 昆仑芯XPU推理
./infer_ppyoloe_demo ./ppyoloe_crn_l_300e_coco 000000014439.jpg 3
# 华为昇腾推理
./infer_ppyoloe_demo ./ppyoloe_crn_l_300e_coco 000000014439.jpg 4
```
以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:

29
examples/vision/detection/paddledetection/cpp/infer_ppyolo.cc
View File

@@ -102,6 +102,33 @@ void GpuInfer(const std::string& model_dir, const std::string& image_file) {
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
auto config_file = model_dir + sep + "infer_cfg.yml";
auto option = fastdeploy::RuntimeOption();
option.UseAscend();
auto model = fastdeploy::vision::detection::PPYOLO(model_file, params_file,
config_file, option);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisDetection(im, res, 0.5);
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
@@ -120,6 +147,8 @@ int main(int argc, char* argv[]) {
GpuInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 2) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

29
examples/vision/detection/paddledetection/cpp/infer_ppyoloe.cc
View File

@@ -131,6 +131,33 @@ void TrtInfer(const std::string& model_dir, const std::string& image_file) {
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
auto config_file = model_dir + sep + "infer_cfg.yml";
auto option = fastdeploy::RuntimeOption();
option.UseAscend();
auto model = fastdeploy::vision::detection::PPYOLOE(model_file, params_file,
config_file, option);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisDetection(im, res, 0.5);
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
@@ -151,6 +178,8 @@ int main(int argc, char* argv[]) {
TrtInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 4) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

29
examples/vision/detection/paddledetection/cpp/infer_ssd.cc
View File

@@ -104,6 +104,33 @@ void GpuInfer(const std::string& model_dir, const std::string& image_file) {
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
auto config_file = model_dir + sep + "infer_cfg.yml";
auto option = fastdeploy::RuntimeOption();
option.UseAscend();
auto model = fastdeploy::vision::detection::SSD(model_file, params_file,
config_file, option);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisDetection(im, res, 0.5);
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
@@ -122,6 +149,8 @@ int main(int argc, char* argv[]) {
GpuInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 2) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

30
examples/vision/detection/paddledetection/cpp/infer_yolov3.cc
View File

@@ -102,6 +102,34 @@ void GpuInfer(const std::string& model_dir, const std::string& image_file) {
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
auto config_file = model_dir + sep + "infer_cfg.yml";
auto option = fastdeploy::RuntimeOption();
option.UseAscend();
auto model = fastdeploy::vision::detection::YOLOv3(model_file, params_file,
config_file, option);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisDetection(im, res, 0.5);
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
@@ -120,6 +148,8 @@ int main(int argc, char* argv[]) {
GpuInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 2) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

2
examples/vision/detection/paddledetection/python/README.md
View File

@@ -25,6 +25,8 @@ python infer_ppyoloe.py --model_dir ppyoloe_crn_l_300e_coco --image 000000014439
python infer_ppyoloe.py --model_dir ppyoloe_crn_l_300e_coco --image 000000014439.jpg --device gpu --use_trt True
# 昆仑芯XPU推理
python infer_ppyoloe.py --model_dir ppyoloe_crn_l_300e_coco --image 000000014439.jpg --device kunlunxin
# 华为昇腾推理
python infer_ppyoloe.py --model_dir ppyoloe_crn_l_300e_coco --image 000000014439.jpg --device ascend
```
运行完成可视化结果如下图所示

3
examples/vision/detection/paddledetection/python/infer_ppyolo.py
View File

@@ -32,6 +32,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
if args.device.lower() == "gpu":
option.use_gpu()

3
examples/vision/detection/paddledetection/python/infer_ppyoloe.py
View File

@@ -33,6 +33,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
if args.device.lower() == "gpu":
option.use_gpu()

3
examples/vision/detection/paddledetection/python/infer_ssd.py
View File

@@ -26,6 +26,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
if args.device.lower() == "gpu":
option.use_gpu()
return option

3
examples/vision/detection/paddledetection/python/infer_yolov3.py
View File

@@ -32,6 +32,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
if args.device.lower() == "gpu":
option.use_gpu()

2
examples/vision/detection/yolov5/cpp/README.md
View File

@@ -31,6 +31,8 @@ wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/0000000
./infer_paddle_demo yolov5s_infer 000000014439.jpg 2
# 昆仑芯XPU推理
./infer_paddle_demo yolov5s_infer 000000014439.jpg 3
# 华为昇腾推理
./infer_paddle_demo yolov5s_infer 000000014439.jpg 4
```
上述的模型为 Paddle 模型的推理，如果想要做 ONNX 模型的推理，可以按照如下步骤：

31
examples/vision/detection/yolov5/cpp/infer_paddle_model.cc
View File

@@ -130,6 +130,35 @@ void KunlunXinInfer(const std::string& model_dir, const std::string& image_file)
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
fastdeploy::RuntimeOption option;
option.UseAscend();
auto model = fastdeploy::vision::detection::YOLOv5(
model_file, params_file, option, fastdeploy::ModelFormat::PADDLE);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisDetection(im, res);
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/image run_option, "
@@ -149,6 +178,8 @@ int main(int argc, char* argv[]) {
TrtInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 4) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

2
examples/vision/detection/yolov5/python/README.md
View File

@@ -25,6 +25,8 @@ python infer.py --model yolov5s_infer --image 000000014439.jpg --device gpu
python infer.py --model yolov5s_infer --image 000000014439.jpg --device gpu --use_trt True
# 昆仑芯XPU推理
python infer.py --model yolov5s_infer --image 000000014439.jpg --device kunlunxin
# 华为昇腾推理
python infer.py --model yolov5s_infer --image 000000014439.jpg --device ascend
```
运行完成可视化结果如下图所示

3
examples/vision/detection/yolov5/python/infer.py
View File

@@ -31,6 +31,9 @@ def build_option(args):
if args.device.lower() == "gpu":
option.use_gpu()
if args.device.lower() == "ascend":
option.use_ascend()
if args.use_trt:
option.use_trt_backend()
option.set_trt_input_shape("images", [1, 3, 640, 640])

2
examples/vision/detection/yolov6/cpp/README.md
View File

@@ -29,6 +29,8 @@ wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/0000000
./infer_paddle_demo yolov6s_infer 000000014439.jpg 1
# 昆仑芯XPU推理
./infer_paddle_demo yolov6s_infer 000000014439.jpg 2
# 华为昇腾推理
./infer_paddle_demo yolov6s_infer 000000014439.jpg 3
```
如果想要验证ONNX模型的推理，可以参考如下命令：

28
examples/vision/detection/yolov6/cpp/infer_paddle_model.cc
View File

@@ -96,6 +96,32 @@ void GpuInfer(const std::string& model_dir, const std::string& image_file) {
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
fastdeploy::RuntimeOption option;
option.UseAscend();
auto model = fastdeploy::vision::detection::YOLOv6(
model_file, params_file, option, fastdeploy::ModelFormat::PADDLE);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(&im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisDetection(im, res);
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) {
@@ -114,6 +140,8 @@ int main(int argc, char* argv[]) {
GpuInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 2) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

2
examples/vision/detection/yolov6/python/README.md
View File

@@ -22,6 +22,8 @@ python infer_paddle_model.py --model yolov6s_infer --image 000000014439.jpg --d
python infer_paddle_model.py --model yolov6s_infer --image 000000014439.jpg --device gpu
# 昆仑芯XPU推理
python infer_paddle_model.py --model yolov6s_infer --image 000000014439.jpg --device kunlunxin
# 华为昇腾推理
python infer_paddle_model.py --model yolov6s_infer --image 000000014439.jpg --device ascend
```
如果想要验证ONNX模型的推理，可以参考如下命令：
```bash

3
examples/vision/detection/yolov6/python/infer_paddle_model.py
View File

@@ -28,6 +28,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
return option

2
examples/vision/detection/yolov7/cpp/README.md
View File

@@ -28,6 +28,8 @@ wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/0000000
./infer_paddle_model_demo yolov7_infer 000000014439.jpg 1
# 昆仑芯XPU推理
./infer_paddle_model_demo yolov7_infer 000000014439.jpg 2
# 华为昇腾推理
./infer_paddle_model_demo yolov7_infer 000000014439.jpg 3
```
如果想要验证ONNX模型的推理，可以参考如下命令：
```bash

4
examples/vision/detection/yolov7/cpp/infer_paddle_model.cc
View File

@@ -31,7 +31,7 @@ void InitAndInfer(const std::string& model_dir, const std::string& image_file,
auto im = cv::imread(image_file);
fastdeploy::vision::DetectionResult res;
if (!model.Predict(&im, &res)) {
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
@@ -68,6 +68,8 @@ int main(int argc, char* argv[]) {
option.UseTrtBackend();
} else if (flag == 2) {
option.UseKunlunXin();
} else if (flag == 3) {
option.UseAscend();
}
std::string model_dir = argv[1];

2
examples/vision/detection/yolov7/python/README.md
View File

@@ -24,6 +24,8 @@ python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --dev
python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --device gpu
# 昆仑芯XPU推理
python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --device kunlunxin
# 华为昇腾推理
python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --device ascend
```
如果想要验证ONNX模型的推理，可以参考如下命令：
```bash

2
examples/vision/detection/yolov7/python/README_EN.md
View File

@@ -24,6 +24,8 @@ python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --dev
python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --device gpu
# KunlunXin XPU
python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --device kunlunxin
# Huawei Ascend
python infer_paddle_model.py --model yolov7_infer --image 000000014439.jpg --device ascend
```
If you want to test ONNX model:
```bash

3
examples/vision/detection/yolov7/python/infer_paddle_model.py
View File

@@ -28,6 +28,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
return option

2
examples/vision/ocr/PP-OCRv2/cpp/README.md
View File

@@ -43,6 +43,8 @@ wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_
./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 3
# 昆仑芯XPU推理
./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 4
# 华为昇腾推理
./infer_demo ./ch_PP-OCRv2_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv2_rec_infer ./ppocr_keys_v1.txt ./12.jpg 5
```
以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:

9
examples/vision/ocr/PP-OCRv2/cpp/infer.cc
View File

@@ -55,6 +55,10 @@ void InitAndInfer(const std::string& det_model_dir, const std::string& cls_model
auto cls_model = fastdeploy::vision::ocr::Classifier(cls_model_file, cls_params_file, cls_option);
auto rec_model = fastdeploy::vision::ocr::Recognizer(rec_model_file, rec_params_file, rec_label_file, rec_option);
// 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);
assert(det_model.Initialized());
assert(cls_model.Initialized());
assert(rec_model.Initialized());
@@ -66,6 +70,9 @@ void InitAndInfer(const std::string& det_model_dir, const std::string& cls_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.
// When users enable static shape infer for rec model, the batch size of cls and rec model needs to be set to 1.
// ppocr_v2.SetClsBatchSize(1);
// ppocr_v2.SetRecBatchSize(1);
ppocr_v2.SetClsBatchSize(cls_batch_size);
ppocr_v2.SetRecBatchSize(rec_batch_size);
@@ -122,6 +129,8 @@ int main(int argc, char* argv[]) {
option.EnablePaddleToTrt();
} else if (flag == 4) {
option.UseKunlunXin();
} else if (flag == 5) {
option.UseAscend();
}
std::string det_model_dir = argv[1];

2
examples/vision/ocr/PP-OCRv2/python/README.md
View File

@@ -36,6 +36,8 @@ python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2
python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv2_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu --backend trt
# 昆仑芯XPU推理
python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv2_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device kunlunxin
# 华为昇腾推理
python infer.py --det_model ch_PP-OCRv2_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv2_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device ascend
```
运行完成可视化结果如下图所示

10
examples/vision/ocr/PP-OCRv2/python/infer.py
View File

@@ -72,6 +72,10 @@ def build_option(args):
option.use_kunlunxin()
return option
if args.device.lower() == "ascend":
option.use_ascend()
return option
if args.backend.lower() == "trt":
assert args.device.lower(
) == "gpu", "TensorRT backend require inference on device GPU."
@@ -112,6 +116,8 @@ runtime_option = build_option(args)
# 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
# 当用户要把PP-OCR部署在对动态shape推理支持有限的设备上时,(例如华为昇腾)
# 需要把cls_batch_size和rec_batch_size都设置为1.
cls_batch_size = 1
rec_batch_size = 6
@@ -144,6 +150,10 @@ rec_option.set_trt_input_shape("x", [1, 3, 32, 10],
rec_model = fd.vision.ocr.Recognizer(
rec_model_file, rec_params_file, rec_label_file, runtime_option=rec_option)
# 当用户要把PP-OCR部署在对动态shape推理支持有限的设备上时,(例如华为昇腾)
# 需要使用下行代码, 来启用rec模型的静态shape推理.
# rec_model.preprocessor.static_shape_infer = True
# 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_v2 = fd.vision.ocr.PPOCRv2(
det_model=det_model, cls_model=cls_model, rec_model=rec_model)

2
examples/vision/ocr/PP-OCRv3/cpp/README.md
View File

@@ -43,6 +43,8 @@ wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_
./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 3
# 昆仑芯XPU推理
./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 4
# 华为昇腾推理, 请用户在代码里正确开启Rec模型的静态shape推理，并设置分类模型和识别模型的推理batch size为1.
./infer_demo ./ch_PP-OCRv3_det_infer ./ch_ppocr_mobile_v2.0_cls_infer ./ch_PP-OCRv3_rec_infer ./ppocr_keys_v1.txt ./12.jpg 5
```
以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:

11
examples/vision/ocr/PP-OCRv3/cpp/infer.cc
View File

@@ -34,7 +34,7 @@ void InitAndInfer(const std::string& det_model_dir, const std::string& cls_model
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.
// User could initialize the inference batch size and set them after create PP-OCR model.
int cls_batch_size = 1;
int rec_batch_size = 6;
@@ -56,6 +56,10 @@ void InitAndInfer(const std::string& det_model_dir, const std::string& cls_model
auto cls_model = fastdeploy::vision::ocr::Classifier(cls_model_file, cls_params_file, cls_option);
auto rec_model = fastdeploy::vision::ocr::Recognizer(rec_model_file, rec_params_file, rec_label_file, rec_option);
// 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);
assert(det_model.Initialized());
assert(cls_model.Initialized());
assert(rec_model.Initialized());
@@ -67,6 +71,9 @@ void InitAndInfer(const std::string& det_model_dir, const std::string& cls_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.
// When users enable static shape infer for rec model, the batch size of cls and rec model needs to be set to 1.
// ppocr_v3.SetClsBatchSize(1);
// ppocr_v3.SetRecBatchSize(1);
ppocr_v3.SetClsBatchSize(cls_batch_size);
ppocr_v3.SetRecBatchSize(rec_batch_size);
@@ -123,6 +130,8 @@ int main(int argc, char* argv[]) {
option.EnablePaddleToTrt();
} else if (flag == 4) {
option.UseKunlunXin();
} else if (flag == 5) {
option.UseAscend();
}
std::string det_model_dir = argv[1];

2
examples/vision/ocr/PP-OCRv3/python/README.md
View File

@@ -35,6 +35,8 @@ python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2
python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu --backend trt
# 昆仑芯XPU推理
python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device kunlunxin
# 华为昇腾推理
python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device ascend
```
运行完成可视化结果如下图所示

10
examples/vision/ocr/PP-OCRv3/python/infer.py
View File

@@ -72,6 +72,10 @@ def build_option(args):
option.use_kunlunxin()
return option
if args.device.lower() == "ascend":
option.use_ascend()
return option
if args.backend.lower() == "trt":
assert args.device.lower(
) == "gpu", "TensorRT backend require inference on device GPU."
@@ -112,6 +116,8 @@ runtime_option = build_option(args)
# 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
# 当用户要把PP-OCR部署在对动态shape推理支持有限的设备上时,(例如华为昇腾)
# 需要把cls_batch_size和rec_batch_size都设置为1.
cls_batch_size = 1
rec_batch_size = 6
@@ -144,6 +150,10 @@ rec_option.set_trt_input_shape("x", [1, 3, 48, 10],
rec_model = fd.vision.ocr.Recognizer(
rec_model_file, rec_params_file, rec_label_file, runtime_option=rec_option)
# 当用户要把PP-OCR部署在对动态shape推理支持有限的设备上时,(例如华为昇腾)
# 需要使用下行代码, 来启用rec模型的静态shape推理.
# rec_model.preprocessor.static_shape_infer = True
# 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)

2
examples/vision/segmentation/paddleseg/cpp/README.md
View File

@@ -34,6 +34,8 @@ wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
./infer_demo Unet_cityscapes_without_argmax_infer cityscapes_demo.png 2
# 昆仑芯XPU推理
./infer_demo Unet_cityscapes_without_argmax_infer cityscapes_demo.png 3
# 华为昇腾推理
./infer_demo Unet_cityscapes_without_argmax_infer cityscapes_demo.png 4
```
运行完成可视化结果如下图所示

30
examples/vision/segmentation/paddleseg/cpp/infer.cc
View File

@@ -135,6 +135,34 @@ void TrtInfer(const std::string& model_dir, const std::string& image_file) {
std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
}
void AscendInfer(const std::string& model_dir, const std::string& image_file) {
auto model_file = model_dir + sep + "model.pdmodel";
auto params_file = model_dir + sep + "model.pdiparams";
auto config_file = model_dir + sep + "deploy.yaml";
auto option = fastdeploy::RuntimeOption();
option.UseAscend();
auto model = fastdeploy::vision::segmentation::PaddleSegModel(
model_file, params_file, config_file, option);
if (!model.Initialized()) {
std::cerr << "Failed to initialize." << std::endl;
return;
}
auto im = cv::imread(image_file);
fastdeploy::vision::SegmentationResult res;
if (!model.Predict(im, &res)) {
std::cerr << "Failed to predict." << std::endl;
return;
}
std::cout << res.Str() << std::endl;
auto vis_im = fastdeploy::vision::VisSegmentation(im, res, 0.5);
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
@@ -155,6 +183,8 @@ int main(int argc, char* argv[]) {
TrtInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 3) {
KunlunXinInfer(argv[1], argv[2]);
} else if (std::atoi(argv[3]) == 4) {
AscendInfer(argv[1], argv[2]);
}
return 0;
}

2
examples/vision/segmentation/paddleseg/python/README.md
View File

@@ -27,6 +27,8 @@ python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_
python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_demo.png --device gpu --use_trt True
# 昆仑芯XPU推理
python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_demo.png --device kunlunxin
# 华为昇腾推理
python infer.py --model Unet_cityscapes_without_argmax_infer --image cityscapes_demo.png --device ascend
```
运行完成可视化结果如下图所示

3
examples/vision/segmentation/paddleseg/python/infer.py
View File

@@ -33,6 +33,9 @@ def build_option(args):
if args.device.lower() == "kunlunxin":
option.use_kunlunxin()
if args.device.lower() == "ascend":
option.use_ascend()
if args.use_trt:
option.use_trt_backend()
option.set_trt_input_shape("x", [1, 3, 256, 256], [1, 3, 1024, 1024],

13
fastdeploy/vision/ocr/ppocr/classifier.h
View File

@@ -68,11 +68,20 @@ class FASTDEPLOY_DECL Classifier : public FastDeployModel {
std::vector<float>* cls_scores,
size_t start_index, size_t end_index);
ClassifierPreprocessor preprocessor_;
ClassifierPostprocessor postprocessor_;
/// Get preprocessor reference of ClassifierPreprocessor
virtual ClassifierPreprocessor& GetPreprocessor() {
return preprocessor_;
}
/// Get postprocessor reference of ClassifierPostprocessor
virtual ClassifierPostprocessor& GetPostprocessor() {
return postprocessor_;
}
private:
bool Initialize();
ClassifierPreprocessor preprocessor_;
ClassifierPostprocessor postprocessor_;
};
} // namespace ocr

6
fastdeploy/vision/ocr/ppocr/cls_postprocessor.h
View File

@@ -39,6 +39,12 @@ class FASTDEPLOY_DECL ClassifierPostprocessor {
std::vector<int32_t>* cls_labels, std::vector<float>* cls_scores,
size_t start_index, size_t total_size);
/// Set threshold for the classification postprocess, default is 0.9
void SetClsThresh(float cls_thresh) { cls_thresh_ = cls_thresh; }
/// Get threshold value of the classification postprocess.
float GetClsThresh() const { return cls_thresh_; }
float cls_thresh_ = 0.9;
};

21
fastdeploy/vision/ocr/ppocr/cls_preprocessor.h
View File

@@ -34,6 +34,27 @@ class FASTDEPLOY_DECL ClassifierPreprocessor {
bool Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs,
size_t start_index, size_t end_index);
/// Set mean value for the image normalization in classification preprocess
void SetMean(std::vector<float> mean) { mean_ = mean; }
/// Get mean value of the image normalization in classification preprocess
std::vector<float> GetMean() const { return mean_; }
/// Set scale value for the image normalization in classification preprocess
void SetScale(std::vector<float> scale) { scale_ = scale; }
/// Get scale value of the image normalization in classification preprocess
std::vector<float> GetScale() const { return scale_; }
/// Set is_scale for the image normalization in classification preprocess
void SetIsScale(bool is_scale) { is_scale_ = is_scale; }
/// Get is_scale of the image normalization in classification preprocess
bool GetIsScale() const { return is_scale_; }
/// Set cls_image_shape for the classification preprocess
void SetClsImageShape(std::vector<int> cls_image_shape)
{ cls_image_shape_ = cls_image_shape; }
/// Get cls_image_shape for the classification preprocess
std::vector<int> GetClsImageShape() const { return cls_image_shape_; }
std::vector<float> mean_ = {0.5f, 0.5f, 0.5f};
std::vector<float> scale_ = {0.5f, 0.5f, 0.5f};
bool is_scale_ = true;

13
fastdeploy/vision/ocr/ppocr/dbdetector.h
View File

@@ -61,11 +61,20 @@ class FASTDEPLOY_DECL DBDetector : public FastDeployModel {
virtual bool BatchPredict(const std::vector<cv::Mat>& images,
std::vector<std::vector<std::array<int, 8>>>* det_results);
DBDetectorPreprocessor preprocessor_;
DBDetectorPostprocessor postprocessor_;
/// Get preprocessor reference of DBDetectorPreprocessor
virtual DBDetectorPreprocessor& GetPreprocessor() {
return preprocessor_;
}
/// Get postprocessor reference of DBDetectorPostprocessor
virtual DBDetectorPostprocessor& GetPostprocessor() {
return postprocessor_;
}
private:
bool Initialize();
DBDetectorPreprocessor preprocessor_;
DBDetectorPostprocessor postprocessor_;
};
} // namespace ocr

28
fastdeploy/vision/ocr/ppocr/det_postprocessor.h
View File

@@ -36,6 +36,34 @@ class FASTDEPLOY_DECL DBDetectorPostprocessor {
std::vector<std::vector<std::array<int, 8>>>* results,
const std::vector<std::array<int, 4>>& batch_det_img_info);
/// Set det_db_thresh for the detection postprocess, default is 0.3
void SetDetDBThresh(double det_db_thresh) { det_db_thresh_ = det_db_thresh; }
/// Get det_db_thresh of the detection postprocess
double GetDetDBThresh() const { return det_db_thresh_; }
/// Set det_db_box_thresh for the detection postprocess, default is 0.6
void SetDetDBBoxThresh(double det_db_box_thresh)
{ det_db_box_thresh_ = det_db_box_thresh; }
/// Get det_db_box_thresh of the detection postprocess
double GetDetDBBoxThresh() const { return det_db_box_thresh_; }
/// Set det_db_unclip_ratio for the detection postprocess, default is 1.5
void SetDetDBUnclipRatio(double det_db_unclip_ratio)
{ det_db_unclip_ratio_ = det_db_unclip_ratio; }
/// Get det_db_unclip_ratio_ of the detection postprocess
double GetDetDBUnclipRatio() const { return det_db_unclip_ratio_; }
/// Set det_db_score_mode for the detection postprocess, default is 'slow'
void SetDetDBScoreMode(std::string det_db_score_mode)
{ det_db_score_mode_ = det_db_score_mode; }
/// Get det_db_score_mode_ of the detection postprocess
std::string GetDetDBScoreMode() const { return det_db_score_mode_; }
/// Set use_dilation for the detection postprocess, default is fasle
void SetUseDilation(int use_dilation) { use_dilation_ = use_dilation; }
/// Get use_dilation of the detection postprocess
int GetUseDilation() const { return use_dilation_; }
double det_db_thresh_ = 0.3;
double det_db_box_thresh_ = 0.6;
double det_db_unclip_ratio_ = 1.5;

20
fastdeploy/vision/ocr/ppocr/det_preprocessor.h
View File

@@ -35,6 +35,26 @@ class FASTDEPLOY_DECL DBDetectorPreprocessor {
std::vector<FDTensor>* outputs,
std::vector<std::array<int, 4>>* batch_det_img_info_ptr);
/// Set max_side_len for the detection preprocess, default is 960
void SetMaxSideLen(int max_side_len) { max_side_len_ = max_side_len; }
/// Get max_side_len of the detection preprocess
int GetMaxSideLen() const { return max_side_len_; }
/// Set mean value for the image normalization in detection preprocess
void SetMean(std::vector<float> mean) { mean_ = mean; }
/// Get mean value of the image normalization in detection preprocess
std::vector<float> GetMean() const { return mean_; }
/// Set scale value for the image normalization in detection preprocess
void SetScale(std::vector<float> scale) { scale_ = scale; }
/// Get scale value of the image normalization in detection preprocess
std::vector<float> GetScale() const { return scale_; }
/// Set is_scale for the image normalization in detection preprocess
void SetIsScale(bool is_scale) { is_scale_ = is_scale; }
/// Get is_scale of the image normalization in detection preprocess
bool GetIsScale() const { return is_scale_; }
int max_side_len_ = 960;
std::vector<float> mean_ = {0.485f, 0.456f, 0.406f};
std::vector<float> scale_ = {0.229f, 0.224f, 0.225f};

51
fastdeploy/vision/ocr/ppocr/ocrmodel_pybind.cc
View File

@@ -24,10 +24,10 @@ void BindPPOCRModel(pybind11::module& m) {
// DBDetector
pybind11::class_<vision::ocr::DBDetectorPreprocessor>(m, "DBDetectorPreprocessor")
.def(pybind11::init<>())
.def_readwrite("max_side_len", &vision::ocr::DBDetectorPreprocessor::max_side_len_)
.def_readwrite("mean", &vision::ocr::DBDetectorPreprocessor::mean_)
.def_readwrite("scale", &vision::ocr::DBDetectorPreprocessor::scale_)
.def_readwrite("is_scale", &vision::ocr::DBDetectorPreprocessor::is_scale_)
.def_property("max_side_len", &vision::ocr::DBDetectorPreprocessor::GetMaxSideLen, &vision::ocr::DBDetectorPreprocessor::SetMaxSideLen)
.def_property("mean", &vision::ocr::DBDetectorPreprocessor::GetMean, &vision::ocr::DBDetectorPreprocessor::SetMean)
.def_property("scale", &vision::ocr::DBDetectorPreprocessor::GetScale, &vision::ocr::DBDetectorPreprocessor::SetScale)
.def_property("is_scale", &vision::ocr::DBDetectorPreprocessor::GetIsScale, &vision::ocr::DBDetectorPreprocessor::SetIsScale)
.def("run", [](vision::ocr::DBDetectorPreprocessor& self, std::vector<pybind11::array>& im_list) {
std::vector<vision::FDMat> images;
for (size_t i = 0; i < im_list.size(); ++i) {
@@ -44,11 +44,12 @@ void BindPPOCRModel(pybind11::module& m) {
pybind11::class_<vision::ocr::DBDetectorPostprocessor>(m, "DBDetectorPostprocessor")
.def(pybind11::init<>())
.def_readwrite("det_db_thresh", &vision::ocr::DBDetectorPostprocessor::det_db_thresh_)
.def_readwrite("det_db_box_thresh", &vision::ocr::DBDetectorPostprocessor::det_db_box_thresh_)
.def_readwrite("det_db_unclip_ratio", &vision::ocr::DBDetectorPostprocessor::det_db_unclip_ratio_)
.def_readwrite("det_db_score_mode", &vision::ocr::DBDetectorPostprocessor::det_db_score_mode_)
.def_readwrite("use_dilation", &vision::ocr::DBDetectorPostprocessor::use_dilation_)
.def_property("det_db_thresh", &vision::ocr::DBDetectorPostprocessor::GetDetDBThresh, &vision::ocr::DBDetectorPostprocessor::SetDetDBThresh)
.def_property("det_db_box_thresh", &vision::ocr::DBDetectorPostprocessor::GetDetDBBoxThresh, &vision::ocr::DBDetectorPostprocessor::SetDetDBBoxThresh)
.def_property("det_db_unclip_ratio", &vision::ocr::DBDetectorPostprocessor::GetDetDBUnclipRatio, &vision::ocr::DBDetectorPostprocessor::SetDetDBUnclipRatio)
.def_property("det_db_score_mode", &vision::ocr::DBDetectorPostprocessor::GetDetDBScoreMode, &vision::ocr::DBDetectorPostprocessor::SetDetDBScoreMode)
.def_property("use_dilation", &vision::ocr::DBDetectorPostprocessor::GetUseDilation, &vision::ocr::DBDetectorPostprocessor::SetUseDilation)
.def("run", [](vision::ocr::DBDetectorPostprocessor& self,
std::vector<FDTensor>& inputs,
const std::vector<std::array<int, 4>>& batch_det_img_info) {
@@ -75,8 +76,8 @@ void BindPPOCRModel(pybind11::module& m) {
.def(pybind11::init<std::string, std::string, RuntimeOption,
ModelFormat>())
.def(pybind11::init<>())
.def_readwrite("preprocessor", &vision::ocr::DBDetector::preprocessor_)
.def_readwrite("postprocessor", &vision::ocr::DBDetector::postprocessor_)
.def_property_readonly("preprocessor", &vision::ocr::DBDetector::GetPreprocessor)
.def_property_readonly("postprocessor", &vision::ocr::DBDetector::GetPostprocessor)
.def("predict", [](vision::ocr::DBDetector& self,
pybind11::array& data) {
auto mat = PyArrayToCvMat(data);
@@ -97,10 +98,10 @@ void BindPPOCRModel(pybind11::module& m) {
// Classifier
pybind11::class_<vision::ocr::ClassifierPreprocessor>(m, "ClassifierPreprocessor")
.def(pybind11::init<>())
.def_readwrite("cls_image_shape", &vision::ocr::ClassifierPreprocessor::cls_image_shape_)
.def_readwrite("mean", &vision::ocr::ClassifierPreprocessor::mean_)
.def_readwrite("scale", &vision::ocr::ClassifierPreprocessor::scale_)
.def_readwrite("is_scale", &vision::ocr::ClassifierPreprocessor::is_scale_)
.def_property("cls_image_shape", &vision::ocr::ClassifierPreprocessor::GetClsImageShape, &vision::ocr::ClassifierPreprocessor::SetClsImageShape)
.def_property("mean", &vision::ocr::ClassifierPreprocessor::GetMean, &vision::ocr::ClassifierPreprocessor::SetMean)
.def_property("scale", &vision::ocr::ClassifierPreprocessor::GetScale, &vision::ocr::ClassifierPreprocessor::SetScale)
.def_property("is_scale", &vision::ocr::ClassifierPreprocessor::GetIsScale, &vision::ocr::ClassifierPreprocessor::SetIsScale)
.def("run", [](vision::ocr::ClassifierPreprocessor& self, std::vector<pybind11::array>& im_list) {
std::vector<vision::FDMat> images;
for (size_t i = 0; i < im_list.size(); ++i) {
@@ -118,7 +119,7 @@ void BindPPOCRModel(pybind11::module& m) {
pybind11::class_<vision::ocr::ClassifierPostprocessor>(m, "ClassifierPostprocessor")
.def(pybind11::init<>())
.def_readwrite("cls_thresh", &vision::ocr::ClassifierPostprocessor::cls_thresh_)
.def_property("cls_thresh", &vision::ocr::ClassifierPostprocessor::GetClsThresh, &vision::ocr::ClassifierPostprocessor::SetClsThresh)
.def("run", [](vision::ocr::ClassifierPostprocessor& self,
std::vector<FDTensor>& inputs) {
std::vector<int> cls_labels;
@@ -144,8 +145,8 @@ void BindPPOCRModel(pybind11::module& m) {
.def(pybind11::init<std::string, std::string, RuntimeOption,
ModelFormat>())
.def(pybind11::init<>())
.def_readwrite("preprocessor", &vision::ocr::Classifier::preprocessor_)
.def_readwrite("postprocessor", &vision::ocr::Classifier::postprocessor_)
.def_property_readonly("preprocessor", &vision::ocr::Classifier::GetPreprocessor)
.def_property_readonly("postprocessor", &vision::ocr::Classifier::GetPostprocessor)
.def("predict", [](vision::ocr::Classifier& self,
pybind11::array& data) {
auto mat = PyArrayToCvMat(data);
@@ -168,11 +169,11 @@ void BindPPOCRModel(pybind11::module& m) {
// Recognizer
pybind11::class_<vision::ocr::RecognizerPreprocessor>(m, "RecognizerPreprocessor")
.def(pybind11::init<>())
.def_readwrite("rec_image_shape", &vision::ocr::RecognizerPreprocessor::rec_image_shape_)
.def_readwrite("mean", &vision::ocr::RecognizerPreprocessor::mean_)
.def_readwrite("scale", &vision::ocr::RecognizerPreprocessor::scale_)
.def_readwrite("is_scale", &vision::ocr::RecognizerPreprocessor::is_scale_)
.def_readwrite("static_shape", &vision::ocr::RecognizerPreprocessor::static_shape_)
.def_property("static_shape_infer", &vision::ocr::RecognizerPreprocessor::GetStaticShapeInfer, &vision::ocr::RecognizerPreprocessor::SetStaticShapeInfer)
.def_property("rec_image_shape", &vision::ocr::RecognizerPreprocessor::GetRecImageShape, &vision::ocr::RecognizerPreprocessor::SetRecImageShape)
.def_property("mean", &vision::ocr::RecognizerPreprocessor::GetMean, &vision::ocr::RecognizerPreprocessor::SetMean)
.def_property("scale", &vision::ocr::RecognizerPreprocessor::GetScale, &vision::ocr::RecognizerPreprocessor::SetScale)
.def_property("is_scale", &vision::ocr::RecognizerPreprocessor::GetIsScale, &vision::ocr::RecognizerPreprocessor::SetIsScale)
.def("run", [](vision::ocr::RecognizerPreprocessor& self, std::vector<pybind11::array>& im_list) {
std::vector<vision::FDMat> images;
for (size_t i = 0; i < im_list.size(); ++i) {
@@ -215,8 +216,8 @@ void BindPPOCRModel(pybind11::module& m) {
.def(pybind11::init<std::string, std::string, std::string, RuntimeOption,
ModelFormat>())
.def(pybind11::init<>())
.def_readwrite("preprocessor", &vision::ocr::Recognizer::preprocessor_)
.def_readwrite("postprocessor", &vision::ocr::Recognizer::postprocessor_)
.def_property_readonly("preprocessor", &vision::ocr::Recognizer::GetPreprocessor)
.def_property_readonly("postprocessor", &vision::ocr::Recognizer::GetPostprocessor)
.def("predict", [](vision::ocr::Recognizer& self,
pybind11::array& data) {
auto mat = PyArrayToCvMat(data);

6
fastdeploy/vision/ocr/ppocr/ppocr_v2.cc
View File

@@ -23,14 +23,14 @@ PPOCRv2::PPOCRv2(fastdeploy::vision::ocr::DBDetector* det_model,
fastdeploy::vision::ocr::Recognizer* rec_model)
: detector_(det_model), classifier_(cls_model), recognizer_(rec_model) {
Initialized();
recognizer_->preprocessor_.rec_image_shape_[1] = 32;
recognizer_->GetPreprocessor().rec_image_shape_[1] = 32;
}
PPOCRv2::PPOCRv2(fastdeploy::vision::ocr::DBDetector* det_model,
fastdeploy::vision::ocr::Recognizer* rec_model)
: detector_(det_model), recognizer_(rec_model) {
Initialized();
recognizer_->preprocessor_.rec_image_shape_[1] = 32;
recognizer_->GetPreprocessor().rec_image_shape_[1] = 32;
}
bool PPOCRv2::SetClsBatchSize(int cls_batch_size) {
@@ -134,7 +134,7 @@ bool PPOCRv2::BatchPredict(const std::vector<cv::Mat>& images,
return false;
}else{
for (size_t i_img = start_index; i_img < end_index; ++i_img) {
if(cls_labels_ptr->at(i_img) % 2 == 1 && cls_scores_ptr->at(i_img) > classifier_->postprocessor_.cls_thresh_) {
if(cls_labels_ptr->at(i_img) % 2 == 1 && cls_scores_ptr->at(i_img) > classifier_->GetPostprocessor().cls_thresh_) {
cv::rotate(image_list[i_img], image_list[i_img], 1);
}
}

4
fastdeploy/vision/ocr/ppocr/ppocr_v3.h
View File

@@ -36,7 +36,7 @@ class FASTDEPLOY_DECL PPOCRv3 : public PPOCRv2 {
fastdeploy::vision::ocr::Recognizer* rec_model)
: PPOCRv2(det_model, cls_model, rec_model) {
// The only difference between v2 and v3
recognizer_->preprocessor_.rec_image_shape_[1] = 48;
recognizer_->GetPreprocessor().rec_image_shape_[1] = 48;
}
/** \brief Classification model is optional, so this function is set up the detection model path and recognition model path respectively.
*
@@ -47,7 +47,7 @@ class FASTDEPLOY_DECL PPOCRv3 : public PPOCRv2 {
fastdeploy::vision::ocr::Recognizer* rec_model)
: PPOCRv2(det_model, rec_model) {
// The only difference between v2 and v3
recognizer_->preprocessor_.rec_image_shape_[1] = 48;
recognizer_->GetPreprocessor().rec_image_shape_[1] = 48;
}
};

23
fastdeploy/vision/ocr/ppocr/rec_preprocessor.cc
View File

@@ -22,12 +22,12 @@ namespace vision {
namespace ocr {
void OcrRecognizerResizeImage(FDMat* mat, float max_wh_ratio,
const std::vector<int>& rec_image_shape, bool static_shape) {
const std::vector<int>& rec_image_shape, bool static_shape_infer) {
int img_h, img_w;
img_h = rec_image_shape[1];
img_w = rec_image_shape[2];
if (!static_shape) {
if (!static_shape_infer) {
img_w = int(img_h * max_wh_ratio);
float ratio = float(mat->Width()) / float(mat->Height());
@@ -52,23 +52,6 @@ void OcrRecognizerResizeImage(FDMat* mat, float max_wh_ratio,
}
}
void OcrRecognizerResizeImageOnAscend(FDMat* mat,
const std::vector<int>& rec_image_shape) {
int img_h, img_w;
img_h = rec_image_shape[1];
img_w = rec_image_shape[2];
if (mat->Width() >= img_w) {
Resize::Run(mat, img_w, img_h); // Reszie W to 320
} else {
Resize::Run(mat, mat->Width(), img_h);
Pad::Run(mat, 0, 0, 0, int(img_w - mat->Width()), {0,0,0});
// Pad to 320
}
}
bool RecognizerPreprocessor::Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs) {
return Run(images, outputs, 0, images->size(), {});
}
@@ -101,7 +84,7 @@ bool RecognizerPreprocessor::Run(std::vector<FDMat>* images, std::vector<FDTenso
real_index = indices[i];
}
FDMat* mat = &(images->at(real_index));
OcrRecognizerResizeImage(mat, max_wh_ratio, rec_image_shape_, static_shape_);
OcrRecognizerResizeImage(mat, max_wh_ratio, rec_image_shape_, static_shape_infer_);
NormalizeAndPermute::Run(mat, mean_, scale_, is_scale_);
}
// Only have 1 output Tensor.

31
fastdeploy/vision/ocr/ppocr/rec_preprocessor.h
View File

@@ -35,11 +35,40 @@ class FASTDEPLOY_DECL RecognizerPreprocessor {
size_t start_index, size_t end_index,
const std::vector<int>& indices);
/// Set static_shape_infer is true or not. When deploy PP-OCR
/// on hardware which can not support dynamic input shape very well,
/// like Huawei Ascned, static_shape_infer needs to to be true.
void SetStaticShapeInfer(bool static_shape_infer)
{ static_shape_infer_ = static_shape_infer; }
/// Get static_shape_infer of the recognition preprocess
bool GetStaticShapeInfer() const { return static_shape_infer_; }
/// Set mean value for the image normalization in recognition preprocess
void SetMean(std::vector<float> mean) { mean_ = mean; }
/// Get mean value of the image normalization in recognition preprocess
std::vector<float> GetMean() const { return mean_; }
/// Set scale value for the image normalization in recognition preprocess
void SetScale(std::vector<float> scale) { scale_ = scale; }
/// Get scale value of the image normalization in recognition preprocess
std::vector<float> GetScale() const { return scale_; }
/// Set is_scale for the image normalization in recognition preprocess
void SetIsScale(bool is_scale) { is_scale_ = is_scale; }
/// Get is_scale of the image normalization in recognition preprocess
bool GetIsScale() const { return is_scale_; }
/// Set rec_image_shape for the recognition preprocess
void SetRecImageShape(std::vector<int> rec_image_shape)
{ rec_image_shape_ = rec_image_shape; }
/// Get rec_image_shape for the recognition preprocess
std::vector<int> GetRecImageShape() const { return rec_image_shape_; }
std::vector<int> rec_image_shape_ = {3, 48, 320};
std::vector<float> mean_ = {0.5f, 0.5f, 0.5f};
std::vector<float> scale_ = {0.5f, 0.5f, 0.5f};
bool is_scale_ = true;
bool static_shape_ = false;
bool static_shape_infer_ = false;
};
} // namespace ocr

13
fastdeploy/vision/ocr/ppocr/recognizer.h
View File

@@ -67,11 +67,20 @@ class FASTDEPLOY_DECL Recognizer : public FastDeployModel {
size_t start_index, size_t end_index,
const std::vector<int>& indices);
RecognizerPreprocessor preprocessor_;
RecognizerPostprocessor postprocessor_;
/// Get preprocessor reference of DBDetectorPreprocessor
virtual RecognizerPreprocessor& GetPreprocessor() {
return preprocessor_;
}
/// Get postprocessor reference of DBDetectorPostprocessor
virtual RecognizerPostprocessor& GetPostprocessor() {
return postprocessor_;
}
private:
bool Initialize();
RecognizerPreprocessor preprocessor_;
RecognizerPostprocessor postprocessor_;
};
} // namespace ocr

24
python/fastdeploy/vision/ocr/ppocr/__init__.py
View File

@@ -509,15 +509,15 @@ class RecognizerPreprocessor:
return self._preprocessor.run(input_ims)
@property
def static_shape(self):
return self._preprocessor.static_shape
def static_shape_infer(self):
return self._preprocessor.static_shape_infer
@static_shape.setter
def static_shape(self, value):
@static_shape_infer.setter
def static_shape_infer(self, value):
assert isinstance(
value,
bool), "The value to set `static_shape` must be type of bool."
self._preprocessor.static_shape = value
bool), "The value to set `static_shape_infer` must be type of bool."
self._preprocessor.static_shape_infer = value
@property
def is_scale(self):
@@ -638,15 +638,15 @@ class Recognizer(FastDeployModel):
self._model.postprocessor = value
@property
def static_shape(self):
return self._model.preprocessor.static_shape
def static_shape_infer(self):
return self._model.preprocessor.static_shape_infer
@static_shape.setter
def static_shape(self, value):
@static_shape_infer.setter
def static_shape_infer(self, value):
assert isinstance(
value,
bool), "The value to set `static_shape` must be type of bool."
self._model.preprocessor.static_shape = value
bool), "The value to set `static_shape_infer` must be type of bool."
self._model.preprocessor.static_shape_infer = value
@property
def is_scale(self):

43
scripts/patch_paddle_lite.py Normal file
View File

@@ -0,0 +1,43 @@
# Copyright (c) 2020 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.
import os
import sys
import shutil
import subprocess
import platform
import sys
def process_paddle_lite(paddle_lite_so_path):
if platform.system().lower() != "linux":
return
rpaths = ["$ORIGIN"]
patchelf_exe = os.getenv("PATCHELF_EXE", "patchelf")
for paddle_lite_so_file in os.listdir(paddle_lite_so_path):
paddle_lite_so_file = os.path.join(paddle_lite_so_path,
paddle_lite_so_file)
if '.so' in paddle_lite_so_file:
command = "{} --set-rpath '{}' {}".format(
patchelf_exe, ":".join(rpaths), paddle_lite_so_file)
if platform.machine() != 'sw_64' and platform.machine(
) != 'mips64':
assert os.system(
command) == 0, "patchelf {} failed, the command: {}".format(
paddle_lite_so_file, command)
if __name__ == "__main__":
process_paddle_lite(sys.argv[1])