mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-11-03 11:02:01 +08:00
[Model] Add picodet for RV1126 and A311D (#1549)
* add infer for picodet * update code * update lite lib --------- Co-authored-by: DefTruth <31974251+DefTruth@users.noreply.github.com>
This commit is contained in:
yeliang2258
@@ -1,12 +1,12 @@
|
||||
English | [简体中文](README_CN.md)
|
||||
# Deploy PP-YOLOE Quantification Model on A311D
|
||||
Now FastDeploy supports the deployment of PP-YOLOE quantification model to A311D on Paddle Lite.
|
||||
Now FastDeploy supports the deployment of PP-YOLOE and PicoDet quantification model to A311D on Paddle Lite.
|
||||
|
||||
For model quantification and download, refer to [Model Quantification](../quantize/README.md)
|
||||
|
||||
|
||||
## Detailed Deployment Tutorials
|
||||
|
||||
Only C++ deployment is supported on A311D
|
||||
Only C++ deployment is supported on A311D
|
||||
|
||||
- [C++ deployment](cpp)
|
||||
|
||||
2
examples/vision/detection/paddledetection/a311d/README_CN.md
Normal file → Executable file
2
examples/vision/detection/paddledetection/a311d/README_CN.md
Normal file → Executable file
@@ -1,6 +1,6 @@
|
||||
[English](README.md) | 简体中文
|
||||
# PP-YOLOE 量化模型在 A311D 上的部署
|
||||
目前 FastDeploy 已经支持基于 Paddle Lite 部署 PP-YOLOE 量化模型到 A311D 上。
|
||||
目前 FastDeploy 已经支持基于 Paddle Lite 部署 PP-YOLOE 和 PicoDet 量化模型到 A311D 上。
|
||||
|
||||
模型的量化和量化模型的下载请参考：[模型量化](../quantize/README.md)
|
||||
|
||||
|
||||
@@ -10,13 +10,16 @@ include(${FASTDEPLOY_INSTALL_DIR}/FastDeploy.cmake)
|
||||
include_directories(${FASTDEPLOY_INCS})
|
||||
include_directories(${FastDeploy_INCLUDE_DIRS})
|
||||
|
||||
add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer_ppyoloe.cc)
|
||||
add_executable(ppyoloe_infer_demo ${PROJECT_SOURCE_DIR}/infer_ppyoloe.cc)
|
||||
add_executable(picodet_infer_demo ${PROJECT_SOURCE_DIR}/infer_picodet.cc)
|
||||
# 添加FastDeploy库依赖
|
||||
target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
|
||||
target_link_libraries(ppyoloe_infer_demo ${FASTDEPLOY_LIBS})
|
||||
target_link_libraries(picodet_infer_demo ${FASTDEPLOY_LIBS})
|
||||
|
||||
set(CMAKE_INSTALL_PREFIX ${CMAKE_SOURCE_DIR}/build/install)
|
||||
|
||||
install(TARGETS infer_demo DESTINATION ./)
|
||||
install(TARGETS ppyoloe_infer_demo DESTINATION ./)
|
||||
install(TARGETS picodet_infer_demo DESTINATION ./)
|
||||
|
||||
install(DIRECTORY models DESTINATION ./)
|
||||
install(DIRECTORY images DESTINATION ./)
|
||||
|
||||
@@ -1,23 +1,23 @@
|
||||
English | [简体中文](README_CN.md)
|
||||
# PP-YOLOE Quantitative Model C++ Deployment Example
|
||||
# PP-YOLOE Quantitative Model C++ Deployment Example
|
||||
|
||||
`infer.cc` in this directory can help you quickly complete the inference acceleration of PP-YOLOE quantization model deployment on A311D.
|
||||
`infer.cc` in this directory can help you quickly complete the inference acceleration of PP-YOLOE and PicoDet quantization model deployment on A311D.
|
||||
|
||||
## Deployment Preparations
|
||||
### FastDeploy Cross-compile Environment Preparations
|
||||
1. For the software and hardware environment, and the cross-compile environment, please refer to [FastDeploy Cross-compile environment](../../../../../../docs/en/build_and_install/a311d.md#Cross-compilation-environment-construction)
|
||||
1. For the software and hardware environment, and the cross-compile environment, please refer to [Preparations for FastDeploy Cross-compile environment](../../../../../../docs/en/build_and_install/a311d.md#Cross-compilation-environment-construction).
|
||||
|
||||
### Model Preparations
|
||||
1. You can directly use the quantized model provided by FastDeploy for deployment.
|
||||
2. You can use PaddleDetection to export Float32 models, note that you need to set the parameter when exporting model: use_shared_conv=False. For more information: [PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe).
|
||||
2. You can use PaddleDetection to export Float32 models, note that you need to set the parameter when exporting PP-YOLOE model: use_shared_conv=False. For more information: [PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe). For more information when exporting PicoDet model: [PicoDet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/picodet).
|
||||
3. You can use [one-click automatical compression tool](../../../../../../tools/common_tools/auto_compression/) provided by FastDeploy to quantize model by yourself, and use the generated quantized model for deployment.(Note: The quantized classification model still needs the infer_cfg.yml file in the FP32 model folder. Self-quantized model folder does not contain this yaml file, you can copy it from the FP32 model folder to the quantized model folder.)
|
||||
4. The model requires heterogeneous computation. Please refer to: [Heterogeneous Computation](./../../../../../../docs/en/faq/heterogeneous_computing_on_timvx_npu.md). Since the model is already provided, you can test the heterogeneous file we provide first to verify whether the accuracy meets the requirements.
|
||||
|
||||
For more information, please refer to [Model Quantization](../../quantize/README.md)
|
||||
|
||||
## Deploying the Quantized PP-YOLOE Detection model on A311D
|
||||
Please follow these steps to complete the deployment of the PP-YOLOE quantization model on A311D.
|
||||
1. Cross-compile the FastDeploy library as described in [Cross-compile FastDeploy](../../../../../../docs/en/build_and_install/a311d.md#FastDeploy-cross-compilation-library-compilation-based-on-Paddle-Lite)
|
||||
## Deploying the Quantized PP-YOLOE and PicoDet Detection model on A311D
|
||||
Please follow these steps to complete the deployment of the PP-YOLOE and PicoDet quantization model on A311D.
|
||||
1. Cross-compile the FastDeploy library as described in [Cross-compile FastDeploy](../../../../../../docs/en/build_and_install/a311d.md#FastDeploy-cross-compilation-library-compilation-based-on-Paddle-Lite)
|
||||
|
||||
2. Copy the compiled library to the current directory. You can run this line:
|
||||
```bash
|
||||
@@ -28,9 +28,14 @@ cp -r FastDeploy/build/fastdeploy-timvx/ FastDeploy/examples/vision/detection/pa
|
||||
```bash
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/a311d/cpp
|
||||
mkdir models && mkdir images
|
||||
# download PP-YOLOE model
|
||||
wget https://bj.bcebos.com/fastdeploy/models/ppyoloe_noshare_qat.tar.gz
|
||||
tar -xvf ppyoloe_noshare_qat.tar.gz
|
||||
cp -r ppyoloe_noshare_qat models
|
||||
# download PicoDet model
|
||||
wget https://bj.bcebos.com/fastdeploy/models/picodet_withNMS_quant_qat.tar.gz
|
||||
tar -xvf picodet_withNMS_quant_qat.tar.gz
|
||||
cp -r picodet_withNMS_quant_qat models
|
||||
wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
|
||||
cp -r 000000014439.jpg images
|
||||
```
|
||||
@@ -45,12 +50,13 @@ make install
|
||||
# After success, an install folder will be created with a running demo and libraries required for deployment.
|
||||
```
|
||||
|
||||
5. Deploy the PP-YOLOE detection model to A311D based on adb.
|
||||
5. Deploy the PP-YOLOE and PicoDet detection model to A311D based on adb. You can run the following lines:
|
||||
```bash
|
||||
# Go to the install directory.
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/a311d/cpp/build/install/
|
||||
# The following line represents: bash run_with_adb.sh, demo needed to run, model path, image path, DEVICE ID.
|
||||
bash run_with_adb.sh infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh ppyoloe_infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh picodet_infer_demo picodet_withNMS_quant_qat 000000014439.jpg $DEVICE_ID
|
||||
```
|
||||
|
||||
The output is:
|
||||
|
||||
19
examples/vision/detection/paddledetection/a311d/cpp/README_CN.md
Normal file → Executable file
19
examples/vision/detection/paddledetection/a311d/cpp/README_CN.md
Normal file → Executable file
@@ -1,7 +1,7 @@
|
||||
[English](README.md) | 简体中文
|
||||
# PP-YOLOE 量化模型 C++ 部署示例
|
||||
|
||||
本目录下提供的 `infer.cc`，可以帮助用户快速完成 PP-YOLOE 量化模型在 A311D 上的部署推理加速。
|
||||
本目录下提供的 `infer.cc`，可以帮助用户快速完成 PP-YOLOE 和 PicoDet 量化模型在 A311D 上的部署推理加速。
|
||||
|
||||
## 部署准备
|
||||
### FastDeploy 交叉编译环境准备
|
||||
@@ -9,14 +9,14 @@
|
||||
|
||||
### 模型准备
|
||||
1. 用户可以直接使用由 FastDeploy 提供的量化模型进行部署。
|
||||
2. 用户可以先使用 PaddleDetection 自行导出 Float32 模型，注意导出模型模型时设置参数：use_shared_conv=False，更多细节请参考：[PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe)
|
||||
2. 用户可以先使用 PaddleDetection 自行导出 Float32 模型，注意导出 PP-YOLOE 模型时设置参数：use_shared_conv=False，更多细节请参考：[PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe)，导出 PicoDet 请参考：[PicoDet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/picodet)
|
||||
3. 用户可以使用 FastDeploy 提供的[一键模型自动化压缩工具](../../../../../../tools/common_tools/auto_compression/)，自行进行模型量化, 并使用产出的量化模型进行部署。（注意: 推理量化后的检测模型仍然需要FP32模型文件夹下的 infer_cfg.yml 文件，自行量化的模型文件夹内不包含此 yaml 文件，用户从 FP32 模型文件夹下复制此yaml文件到量化后的模型文件夹内即可。）
|
||||
4. 模型需要异构计算，异构计算文件可以参考：[异构计算](./../../../../../../docs/cn/faq/heterogeneous_computing_on_timvx_npu.md)，由于 FastDeploy 已经提供了模型，可以先测试我们提供的异构文件，验证精度是否符合要求。
|
||||
|
||||
更多量化相关相关信息可查阅[模型量化](../../quantize/README.md)
|
||||
|
||||
## 在 A311D 上部署量化后的 PP-YOLOE 检测模型
|
||||
请按照以下步骤完成在 A311D 上部署 PP-YOLOE 量化模型：
|
||||
## 在 A311D 上部署量化后的 PP-YOLOE 和 PicoDet 检测模型
|
||||
请按照以下步骤完成在 A311D 上部署 PP-YOLOE 和 PicoDet 量化模型：
|
||||
1. 交叉编译编译 FastDeploy 库，具体请参考：[交叉编译 FastDeploy](../../../../../../docs/cn/build_and_install/a311d.md#基于-paddlelite-的-fastdeploy-交叉编译库编译)
|
||||
|
||||
2. 将编译后的库拷贝到当前目录，可使用如下命令：
|
||||
@@ -28,9 +28,15 @@ cp -r FastDeploy/build/fastdeploy-timvx/ FastDeploy/examples/vision/detection/pa
|
||||
```bash
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/a311d/cpp
|
||||
mkdir models && mkdir images
|
||||
# 下载 FastDeploy 准备的 PP-YOLOE 模型
|
||||
wget https://bj.bcebos.com/fastdeploy/models/ppyoloe_noshare_qat.tar.gz
|
||||
tar -xvf ppyoloe_noshare_qat.tar.gz
|
||||
cp -r ppyoloe_noshare_qat models
|
||||
# 下载 FastDeploy 准备的 PicoDet 模型
|
||||
wget https://bj.bcebos.com/fastdeploy/models/picodet_withNMS_quant_qat.tar.gz
|
||||
tar -xvf picodet_withNMS_quant_qat.tar.gz
|
||||
cp -r picodet_withNMS_quant_qat models
|
||||
# 下载 FastDeploy 准备的测试图片
|
||||
wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
|
||||
cp -r 000000014439.jpg images
|
||||
```
|
||||
@@ -45,12 +51,13 @@ make install
|
||||
# 成功编译之后，会生成 install 文件夹，里面有一个运行 demo 和部署所需的库
|
||||
```
|
||||
|
||||
5. 基于 adb 工具部署 PP-YOLOE 检测模型到晶晨 A311D
|
||||
5. 基于 adb 工具部署 PP-YOLOE 和 PicoDet 检测模型到 A311D，可使用如下命令：
|
||||
```bash
|
||||
# 进入 install 目录
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/a311d/cpp/build/install/
|
||||
# 如下命令表示：bash run_with_adb.sh 需要运行的demo 模型路径 图片路径 设备的DEVICE_ID
|
||||
bash run_with_adb.sh infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh ppyoloe_infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh picodet_infer_demo picodet_withNMS_quant_qat 000000014439.jpg $DEVICE_ID
|
||||
```
|
||||
|
||||
部署成功后运行结果如下：
|
||||
|
||||
64
examples/vision/detection/paddledetection/a311d/cpp/infer_picodet.cc
Executable file
64
examples/vision/detection/paddledetection/a311d/cpp/infer_picodet.cc
Executable file
@@ -0,0 +1,64 @@
|
||||
// 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& 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 subgraph_file = model_dir + sep + "subgraph.txt";
|
||||
fastdeploy::vision::EnableFlyCV();
|
||||
fastdeploy::RuntimeOption option;
|
||||
option.UseTimVX();
|
||||
option.SetLiteSubgraphPartitionPath(subgraph_file);
|
||||
|
||||
auto model = fastdeploy::vision::detection::PicoDet(model_file, params_file,
|
||||
config_file, option);
|
||||
assert(model.Initialized());
|
||||
|
||||
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, 0.3);
|
||||
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 < 3) {
|
||||
std::cout << "Usage: picodet_infer_demo path/to/quant_model "
|
||||
"path/to/image "
|
||||
"e.g ./picodet_infer_demo ./picodet_ptq ./test.jpeg"
|
||||
<< std::endl;
|
||||
return -1;
|
||||
}
|
||||
|
||||
std::string model_dir = argv[1];
|
||||
std::string test_image = argv[2];
|
||||
InitAndInfer(model_dir, test_image);
|
||||
return 0;
|
||||
}
|
||||
@@ -24,7 +24,7 @@ void InitAndInfer(const std::string& model_dir, const std::string& image_file) {
|
||||
auto params_file = model_dir + sep + "model.pdiparams";
|
||||
auto config_file = model_dir + sep + "infer_cfg.yml";
|
||||
auto subgraph_file = model_dir + sep + "subgraph.txt";
|
||||
fastdeploy::vision::EnableFlyCV();
|
||||
fastdeploy::vision::EnableFlyCV();
|
||||
fastdeploy::RuntimeOption option;
|
||||
option.UseTimVX();
|
||||
option.SetLiteSubgraphPartitionPath(subgraph_file);
|
||||
@@ -46,14 +46,13 @@ void InitAndInfer(const std::string& model_dir, const std::string& image_file) {
|
||||
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 < 3) {
|
||||
std::cout << "Usage: infer_demo path/to/quant_model "
|
||||
std::cout << "Usage: ppyoloe_infer_demo path/to/quant_model "
|
||||
"path/to/image "
|
||||
"e.g ./infer_demo ./PPYOLOE_L_quant ./test.jpeg"
|
||||
"e.g ./ppyoloe_infer_demo ./PPYOLOE_L_quant ./test.jpeg"
|
||||
<< std::endl;
|
||||
return -1;
|
||||
}
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
English | [简体中文](README_CN.md)
|
||||
# Deploy PP-YOLOE Quantification Model on RV1126
|
||||
Now FastDeploy supports the deployment of PP-YOLOE quantification model to RV1126 based on Paddle Lite.
|
||||
# Deploy PP-YOLOE and PicoDet Quantification Model on RV1126
|
||||
Now FastDeploy supports the deployment of PP-YOLOE and PicoDet quantification model to RV1126 based on Paddle Lite.
|
||||
|
||||
For model quantification and download, refer to [Model Quantification](../quantize/README.md)
|
||||
|
||||
|
||||
4
examples/vision/detection/paddledetection/rv1126/README_CN.md
Normal file → Executable file
4
examples/vision/detection/paddledetection/rv1126/README_CN.md
Normal file → Executable file
@@ -1,6 +1,6 @@
|
||||
[English](README.md) | 简体中文
|
||||
# PP-YOLOE 量化模型在 RV1126 上的部署
|
||||
目前 FastDeploy 已经支持基于 Paddle Lite 部署 PP-YOLOE 量化模型到 RV1126 上。
|
||||
# PP-YOLOE 和 PicoDet 量化模型在 RV1126 上的部署
|
||||
目前 FastDeploy 已经支持基于 Paddle Lite 部署 PP-YOLOE 和 PicoDet 量化模型到 RV1126 上。
|
||||
|
||||
模型的量化和量化模型的下载请参考：[模型量化](../quantize/README.md)
|
||||
|
||||
|
||||
@@ -10,13 +10,16 @@ include(${FASTDEPLOY_INSTALL_DIR}/FastDeploy.cmake)
|
||||
include_directories(${FASTDEPLOY_INCS})
|
||||
include_directories(${FastDeploy_INCLUDE_DIRS})
|
||||
|
||||
add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer_ppyoloe.cc)
|
||||
add_executable(ppyoloe_infer_demo ${PROJECT_SOURCE_DIR}/infer_ppyoloe.cc)
|
||||
add_executable(picodet_infer_demo ${PROJECT_SOURCE_DIR}/infer_picodet.cc)
|
||||
# 添加FastDeploy库依赖
|
||||
target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
|
||||
target_link_libraries(ppyoloe_infer_demo ${FASTDEPLOY_LIBS})
|
||||
target_link_libraries(picodet_infer_demo ${FASTDEPLOY_LIBS})
|
||||
|
||||
set(CMAKE_INSTALL_PREFIX ${CMAKE_SOURCE_DIR}/build/install)
|
||||
|
||||
install(TARGETS infer_demo DESTINATION ./)
|
||||
install(TARGETS ppyoloe_infer_demo DESTINATION ./)
|
||||
install(TARGETS picodet_infer_demo DESTINATION ./)
|
||||
|
||||
install(DIRECTORY models DESTINATION ./)
|
||||
install(DIRECTORY images DESTINATION ./)
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
English | [简体中文](README_CN.md)
|
||||
# PP-YOLOE Quantitative Model C++ Deployment Example
|
||||
|
||||
`infer.cc` in this directory can help you quickly complete the inference acceleration of PP-YOLOE quantization model deployment on RV1126.
|
||||
`infer.cc` in this directory can help you quickly complete the inference acceleration of PP-YOLOE and PicoDet quantization model deployment on RV1126.
|
||||
|
||||
## Deployment Preparations
|
||||
### FastDeploy Cross-compile Environment Preparations
|
||||
@@ -9,14 +9,14 @@ English | [简体中文](README_CN.md)
|
||||
|
||||
### Model Preparations
|
||||
1. You can directly use the quantized model provided by FastDeploy for deployment.
|
||||
2. You can use PaddleDetection to export Float32 models, note that you need to set the parameter when exporting model: use_shared_conv=False. For more information: [PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe).
|
||||
2. You can use PaddleDetection to export Float32 models, note that you need to set the parameter when exporting PP-YOLOE model: use_shared_conv=False. For more information: [PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe). For more information when exporting PicoDet model: [PicoDet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/picodet).
|
||||
3. You can use [one-click automatical compression tool](../../../../../../tools/common_tools/auto_compression/) provided by FastDeploy to quantize model by yourself, and use the generated quantized model for deployment.(Note: The quantized classification model still needs the infer_cfg.yml file in the FP32 model folder. Self-quantized model folder does not contain this yaml file, you can copy it from the FP32 model folder to the quantized model folder.)
|
||||
4. The model requires heterogeneous computation. Please refer to: [Heterogeneous Computation](./../../../../../../docs/en/faq/heterogeneous_computing_on_timvx_npu.md). Since the model is already provided, you can test the heterogeneous file we provide first to verify whether the accuracy meets the requirements.
|
||||
|
||||
For more information, please refer to [Model Quantization](../../quantize/README.md)
|
||||
|
||||
## Deploying the Quantized PP-YOLOE Detection model on RV1126
|
||||
Please follow these steps to complete the deployment of the PP-YOLOE quantization model on RV1126.
|
||||
## Deploying the Quantized PP-YOLOE and PicoDet Detection model on RV1126
|
||||
Please follow these steps to complete the deployment of the PP-YOLOE and PicoDet quantization model on RV1126.
|
||||
1. Cross-compile the FastDeploy library as described in [Cross-compile FastDeploy](../../../../../../docs/en/build_and_install/rv1126.md#FastDeploy-cross-compilation-library-compilation-based-on-Paddle-Lite)
|
||||
|
||||
2. Copy the compiled library to the current directory. You can run this line:
|
||||
@@ -28,9 +28,14 @@ cp -r FastDeploy/build/fastdeploy-timvx/ FastDeploy/examples/vision/detection/pa
|
||||
```bash
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/rv1126/cpp
|
||||
mkdir models && mkdir images
|
||||
# download PP-YOLOE model
|
||||
wget https://bj.bcebos.com/fastdeploy/models/ppyoloe_noshare_qat.tar.gz
|
||||
tar -xvf ppyoloe_noshare_qat.tar.gz
|
||||
cp -r ppyoloe_noshare_qat models
|
||||
# download PicoDet model
|
||||
wget https://bj.bcebos.com/fastdeploy/models/picodet_withNMS_quant_qat.tar.gz
|
||||
tar -xvf picodet_withNMS_quant_qat.tar.gz
|
||||
cp -r picodet_withNMS_quant_qat models
|
||||
wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
|
||||
cp -r 000000014439.jpg images
|
||||
```
|
||||
@@ -45,12 +50,13 @@ make install
|
||||
# After success, an install folder will be created with a running demo and libraries required for deployment.
|
||||
```
|
||||
|
||||
5. Deploy the PP-YOLOE detection model to Rockchip RV1126 based on adb. You can run the following lines:
|
||||
5. Deploy the PP-YOLOE and PicoDet detection model to Rockchip RV1126 based on adb. You can run the following lines:
|
||||
```bash
|
||||
# Go to the install directory.
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/rv1126/cpp/build/install/
|
||||
# The following line represents: bash run_with_adb.sh, demo needed to run, model path, image path, DEVICE ID.
|
||||
bash run_with_adb.sh infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh ppyoloe_infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh picodet_infer_demo picodet_withNMS_quant_qat 000000014439.jpg $DEVICE_ID
|
||||
```
|
||||
|
||||
The output is:
|
||||
|
||||
19
examples/vision/detection/paddledetection/rv1126/cpp/README_CN.md
Normal file → Executable file
19
examples/vision/detection/paddledetection/rv1126/cpp/README_CN.md
Normal file → Executable file
@@ -1,7 +1,7 @@
|
||||
[English](README.md) | 简体中文
|
||||
# PP-YOLOE 量化模型 C++ 部署示例
|
||||
|
||||
本目录下提供的 `infer.cc`，可以帮助用户快速完成 PP-YOLOE 量化模型在 RV1126 上的部署推理加速。
|
||||
本目录下提供的 `infer.cc`，可以帮助用户快速完成 PP-YOLOE 和 PicoDet 量化模型在 RV1126 上的部署推理加速。
|
||||
|
||||
## 部署准备
|
||||
### FastDeploy 交叉编译环境准备
|
||||
@@ -9,14 +9,14 @@
|
||||
|
||||
### 模型准备
|
||||
1. 用户可以直接使用由 FastDeploy 提供的量化模型进行部署。
|
||||
2. 用户可以先使用 PaddleDetection 自行导出 Float32 模型，注意导出模型模型时设置参数：use_shared_conv=False，更多细节请参考：[PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe)
|
||||
2. 用户可以先使用 PaddleDetection 自行导出 Float32 模型，注意导出 PP-YOLOE 模型时设置参数：use_shared_conv=False，更多细节请参考：[PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe)，导出 PicoDet 请参考：[PicoDet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/picodet)
|
||||
3. 用户可以使用 FastDeploy 提供的[一键模型自动化压缩工具](../../../../../../tools/common_tools/auto_compression/)，自行进行模型量化, 并使用产出的量化模型进行部署。（注意: 推理量化后的检测模型仍然需要FP32模型文件夹下的 infer_cfg.yml 文件，自行量化的模型文件夹内不包含此 yaml 文件，用户从 FP32 模型文件夹下复制此yaml文件到量化后的模型文件夹内即可。）
|
||||
4. 模型需要异构计算，异构计算文件可以参考：[异构计算](./../../../../../../docs/cn/faq/heterogeneous_computing_on_timvx_npu.md)，由于 FastDeploy 已经提供了模型，可以先测试我们提供的异构文件，验证精度是否符合要求。
|
||||
|
||||
更多量化相关相关信息可查阅[模型量化](../../quantize/README.md)
|
||||
|
||||
## 在 RV1126 上部署量化后的 PP-YOLOE 检测模型
|
||||
请按照以下步骤完成在 RV1126 上部署 PP-YOLOE 量化模型：
|
||||
## 在 RV1126 上部署量化后的 PP-YOLOE 和 PicoDet 检测模型
|
||||
请按照以下步骤完成在 RV1126 上部署 PP-YOLOE 和 PicoDet 量化模型：
|
||||
1. 交叉编译编译 FastDeploy 库，具体请参考：[交叉编译 FastDeploy](../../../../../../docs/cn/build_and_install/rv1126.md#基于-paddlelite-的-fastdeploy-交叉编译库编译)
|
||||
|
||||
2. 将编译后的库拷贝到当前目录，可使用如下命令：
|
||||
@@ -28,9 +28,15 @@ cp -r FastDeploy/build/fastdeploy-timvx/ FastDeploy/examples/vision/detection/pa
|
||||
```bash
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/rv1126/cpp
|
||||
mkdir models && mkdir images
|
||||
# 下载 FastDeploy 准备的 PP-YOLOE 模型
|
||||
wget https://bj.bcebos.com/fastdeploy/models/ppyoloe_noshare_qat.tar.gz
|
||||
tar -xvf ppyoloe_noshare_qat.tar.gz
|
||||
cp -r ppyoloe_noshare_qat models
|
||||
# 下载 FastDeploy 准备的 PicoDet 模型
|
||||
wget https://bj.bcebos.com/fastdeploy/models/picodet_withNMS_quant_qat.tar.gz
|
||||
tar -xvf picodet_withNMS_quant_qat.tar.gz
|
||||
cp -r picodet_withNMS_quant_qat models
|
||||
# 下载 FastDeploy 准备的测试图片
|
||||
wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
|
||||
cp -r 000000014439.jpg images
|
||||
```
|
||||
@@ -45,12 +51,13 @@ make install
|
||||
# 成功编译之后，会生成 install 文件夹，里面有一个运行 demo 和部署所需的库
|
||||
```
|
||||
|
||||
5. 基于 adb 工具部署 PP-YOLOE 检测模型到 Rockchip RV1126，可使用如下命令：
|
||||
5. 基于 adb 工具部署 PP-YOLOE 和 PicoDet 检测模型到 Rockchip RV1126，可使用如下命令：
|
||||
```bash
|
||||
# 进入 install 目录
|
||||
cd FastDeploy/examples/vision/detection/paddledetection/rv1126/cpp/build/install/
|
||||
# 如下命令表示：bash run_with_adb.sh 需要运行的demo 模型路径 图片路径 设备的DEVICE_ID
|
||||
bash run_with_adb.sh infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh ppyoloe_infer_demo ppyoloe_noshare_qat 000000014439.jpg $DEVICE_ID
|
||||
bash run_with_adb.sh picodet_infer_demo picodet_withNMS_quant_qat 000000014439.jpg $DEVICE_ID
|
||||
```
|
||||
|
||||
部署成功后运行结果如下：
|
||||
|
||||
64
examples/vision/detection/paddledetection/rv1126/cpp/infer_picodet.cc
Executable file
64
examples/vision/detection/paddledetection/rv1126/cpp/infer_picodet.cc
Executable file
@@ -0,0 +1,64 @@
|
||||
// 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& 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 subgraph_file = model_dir + sep + "subgraph.txt";
|
||||
fastdeploy::vision::EnableFlyCV();
|
||||
fastdeploy::RuntimeOption option;
|
||||
option.UseTimVX();
|
||||
option.SetLiteSubgraphPartitionPath(subgraph_file);
|
||||
|
||||
auto model = fastdeploy::vision::detection::PicoDet(model_file, params_file,
|
||||
config_file, option);
|
||||
assert(model.Initialized());
|
||||
|
||||
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, 0.3);
|
||||
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 < 3) {
|
||||
std::cout << "Usage: picodet_infer_demo path/to/quant_model "
|
||||
"path/to/image "
|
||||
"e.g ./picodet_infer_demo ./picodet_ptq ./test.jpeg"
|
||||
<< std::endl;
|
||||
return -1;
|
||||
}
|
||||
|
||||
std::string model_dir = argv[1];
|
||||
std::string test_image = argv[2];
|
||||
InitAndInfer(model_dir, test_image);
|
||||
return 0;
|
||||
}
|
||||
7
examples/vision/detection/paddledetection/rv1126/cpp/infer_ppyoloe.cc
Executable file → Normal file
7
examples/vision/detection/paddledetection/rv1126/cpp/infer_ppyoloe.cc
Executable file → Normal file
@@ -24,7 +24,7 @@ void InitAndInfer(const std::string& model_dir, const std::string& image_file) {
|
||||
auto params_file = model_dir + sep + "model.pdiparams";
|
||||
auto config_file = model_dir + sep + "infer_cfg.yml";
|
||||
auto subgraph_file = model_dir + sep + "subgraph.txt";
|
||||
fastdeploy::vision::EnableFlyCV();
|
||||
fastdeploy::vision::EnableFlyCV();
|
||||
fastdeploy::RuntimeOption option;
|
||||
option.UseTimVX();
|
||||
option.SetLiteSubgraphPartitionPath(subgraph_file);
|
||||
@@ -46,14 +46,13 @@ void InitAndInfer(const std::string& model_dir, const std::string& image_file) {
|
||||
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 < 3) {
|
||||
std::cout << "Usage: infer_demo path/to/quant_model "
|
||||
std::cout << "Usage: ppyoloe_infer_demo path/to/quant_model "
|
||||
"path/to/image "
|
||||
"e.g ./infer_demo ./PPYOLOE_L_quant ./test.jpeg"
|
||||
"e.g ./ppyoloe_infer_demo ./PPYOLOE_L_quant ./test.jpeg"
|
||||
<< std::endl;
|
||||
return -1;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user