[Backend] Add YOLOv5、PPYOLOE and PP-Liteseg for RV1126 (#647)

* add yolov5 and ppyoloe for rk1126 * update code, rename rk1126 to rv1126 * add PP-Liteseg * update lite lib * updade doc for PPYOLOE * update doc * fix docs * fix doc and examples * update code * uodate doc * update doc Co-authored-by: Jason <jiangjiajun@baidu.com>
2025-10-06 00:57:33 +08:00 · 2022-12-05 16:48:00 +08:00
parent 763da985f3
commit 104d965b38
59 changed files with 804 additions and 987 deletions
--- a/README_CN.md
+++ b/README_CN.md
@@ -68,8 +68,8 @@

 <div id="fastdeploy-quick-start-python"></div>

-<details close> 
-    
+<details close>
+
 <summary><b>Python SDK快速开始（点开查看详情）</b></summary><div>

 #### 快速安装
@@ -131,7 +131,7 @@ cv2.imwrite("vis_image.jpg", vis_im)
 <div id="fastdeploy-quick-start-cpp"></div>

 <details close>
-    
+
 <summary><b>C++ SDK快速开始（点开查看详情）</b></summary><div>


--- a/README_EN.md
+++ b/README_EN.md
--- a/cmake/opencv.cmake
+++ b/cmake/opencv.cmake
@@ -41,10 +41,12 @@ elseif(IOS)
 else()
  if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
    set(OPENCV_FILENAME "opencv-linux-aarch64-3.4.14")
-  elseif(TARGET_ABI MATCHES "armhf")
-    set(OPENCV_FILENAME "opencv-armv7hf")
  else()
-    set(OPENCV_FILENAME "opencv-linux-x64-3.4.16")
+    if(ENABLE_TIMVX)
+      set(OPENCV_FILENAME "opencv-armv7hf")
+    else()
+      set(OPENCV_FILENAME "opencv-linux-x64-3.4.16")
+    endif()
  endif()
  if(ENABLE_OPENCV_CUDA)
    if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
@@ -57,7 +59,7 @@ endif()
 set(OPENCV_INSTALL_DIR ${THIRD_PARTY_PATH}/install/)
 if(ANDROID)
  set(OPENCV_URL_PREFIX "https://bj.bcebos.com/fastdeploy/third_libs")
-elseif(TARGET_ABI MATCHES "armhf")
+elseif(ENABLE_TIMVX)
  set(OPENCV_URL_PREFIX "https://bj.bcebos.com/fastdeploy/test")
 else() # TODO: use fastdeploy/third_libs instead.
  set(OPENCV_URL_PREFIX "https://bj.bcebos.com/paddle2onnx/libs")
@@ -185,7 +187,7 @@ else()
    file(RENAME ${THIRD_PARTY_PATH}/install/${OPENCV_FILENAME}/ ${THIRD_PARTY_PATH}/install/opencv)
    set(OPENCV_FILENAME opencv)
    set(OpenCV_DIR ${THIRD_PARTY_PATH}/install/${OPENCV_FILENAME})
-    if(TARGET_ABI MATCHES "armhf")
+    if(ENABLE_TIMVX)
      set(OpenCV_DIR ${OpenCV_DIR}/lib/cmake/opencv4)
    endif()
    if (WIN32)
--- a/cmake/paddlelite.cmake
+++ b/cmake/paddlelite.cmake
@@ -59,10 +59,12 @@ elseif(ANDROID)
 else() # Linux
  if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
    set(PADDLELITE_URL "${PADDLELITE_URL_PREFIX}/lite-linux-arm64-20220920.tgz")
-  elseif(TARGET_ABI MATCHES "armhf")
-    set(PADDLELITE_URL "https://bj.bcebos.com/fastdeploy/test/lite-linux_armhf_1101.tgz")
  else()
-    message(FATAL_ERROR "Only support Linux aarch64 now, x64 is not supported with backend Paddle Lite.")
+    if(ENABLE_TIMVX)
+      set(PADDLELITE_URL "https://bj.bcebos.com/fastdeploy/test/lite-linux_armhf_1130.tgz")
+    else()
+      message(FATAL_ERROR "Only support Linux aarch64 or ENABLE_TIMVX now, x64 is not supported with backend Paddle Lite.")
+    endif()
  endif()
 endif()

--- a/cmake/timvx.cmake
+++ b/cmake/timvx.cmake
@@ -1,11 +1,10 @@
-if (NOT DEFINED TARGET_ABI)
+if (NOT DEFINED CMAKE_SYSTEM_PROCESSOR)
    set(CMAKE_SYSTEM_NAME Linux)
    set(CMAKE_SYSTEM_PROCESSOR arm)
    set(CMAKE_C_COMPILER "arm-linux-gnueabihf-gcc")
    set(CMAKE_CXX_COMPILER "arm-linux-gnueabihf-g++")
    set(CMAKE_CXX_FLAGS "-march=armv7-a -mfloat-abi=hard -mfpu=neon-vfpv4 ${CMAKE_CXX_FLAGS}")
    set(CMAKE_C_FLAGS "-march=armv7-a -mfloat-abi=hard -mfpu=neon-vfpv4 ${CMAKE_C_FLAGS}" )
-    set(TARGET_ABI armhf)
    set(CMAKE_BUILD_TYPE MinSizeRel)
 else()
    if(NOT ${ENABLE_LITE_BACKEND})
--- a/docs/cn/build_and_install/README.md
+++ b/docs/cn/build_and_install/README.md
@@ -8,10 +8,10 @@
 ## 自行编译安装
 - [GPU部署环境](gpu.md)
 - [CPU部署环境](cpu.md)
- [CPU部署环境](ipu.md)
+- [IPU部署环境](ipu.md)
 - [Jetson部署环境](jetson.md)
 - [Android平台部署环境](android.md)
- [瑞芯微RK1126部署环境](rk1126.md)
+- [瑞芯微RV1126部署环境](rv1126.md)


 ## FastDeploy编译选项说明
@@ -22,6 +22,7 @@
 | ENABLE_PADDLE_BACKEND   | 默认OFF，是否编译集成Paddle Inference后端(CPU/GPU上推荐打开)                              |  
 | ENABLE_LITE_BACKEND     | 默认OFF，是否编译集成Paddle Lite后端(编译Android库时需要设置为ON)                             |
 | ENABLE_RKNPU2_BACKEND   | 默认OFF，是否编译集成RKNPU2后端(RK3588/RK3568/RK3566上推荐打开)                           |
+| ENABLE_TIMVX            | 默认OFF，需要在RV1126/RV1109上部署时，需设置为ON                                          |
 | ENABLE_TRT_BACKEND      | 默认OFF，是否编译集成TensorRT后端(GPU上推荐打开)                                          |
 | ENABLE_OPENVINO_BACKEND | 默认OFF，是否编译集成OpenVINO后端(CPU上推荐打开)                                          |
 | ENABLE_VISION           | 默认OFF，是否编译集成视觉模型的部署模块                                                     |
--- a/docs/cn/build_and_install/rk1126.md
+++ b/docs/cn/build_and_install/rk1126.md
@@ -1,63 +0,0 @@
-# 瑞芯微 RK1126 部署环境编译安装
-
-FastDeploy基于 Paddle-Lite 后端支持在瑞芯微（Rockchip）Soc 上进行部署推理。
-更多详细的信息请参考：[PaddleLite部署示例](https://paddle-lite.readthedocs.io/zh/develop/demo_guides/verisilicon_timvx.html)。
-
-本文档介绍如何编译基于 PaddleLite 的 C++ FastDeploy 交叉编译库。
-
-相关编译选项说明如下：  
-|编译选项|默认值|说明|备注|  
-|:---|:---|:---|:---|  
-|ENABLE_LITE_BACKEND|OFF|编译RK库时需要设置为ON| - |
-
-更多编译选项请参考[FastDeploy编译选项说明](./README.md)
-
-## 交叉编译环境搭建
-
-### 宿主机环境需求  
- os：Ubuntu == 16.04
- cmake： version >= 3.10.0  
-
-### 环境搭建
-```bash
- # 1. Install basic software
-apt update
-apt-get install -y --no-install-recommends \
-  gcc g++ git make wget python unzip
-
-# 2. Install arm gcc toolchains
-apt-get install -y --no-install-recommends \
-  g++-arm-linux-gnueabi gcc-arm-linux-gnueabi \
-  g++-arm-linux-gnueabihf gcc-arm-linux-gnueabihf \
-  gcc-aarch64-linux-gnu g++-aarch64-linux-gnu
-
-# 3. Install cmake 3.10 or above
-wget -c https://mms-res.cdn.bcebos.com/cmake-3.10.3-Linux-x86_64.tar.gz && \
-  tar xzf cmake-3.10.3-Linux-x86_64.tar.gz && \
-  mv cmake-3.10.3-Linux-x86_64 /opt/cmake-3.10 && \
-  ln -s /opt/cmake-3.10/bin/cmake /usr/bin/cmake && \
-  ln -s /opt/cmake-3.10/bin/ccmake /usr/bin/ccmake
-```
-
-## 基于 PaddleLite 的 FastDeploy 交叉编译库编译
-搭建好交叉编译环境之后，编译命令如下：
-```bash
-# Download the latest source code
-git clone https://github.com/PaddlePaddle/FastDeploy.git
-cd FastDeploy  
-mkdir build && cd build
-
-# CMake configuration with RK toolchain
-cmake -DCMAKE_TOOLCHAIN_FILE=./../cmake/timvx.cmake \
-      -DENABLE_TIMVX=ON  \
-      -DCMAKE_INSTALL_PREFIX=fastdeploy-tmivx \
-      -DENABLE_VISION=ON \ # 是否编译集成视觉模型的部署模块，可选择开启
-      -Wno-dev ..
-
-# Build FastDeploy RK1126 C++ SDK
-make -j8
-make install
-```  
-编译完成之后，会生成 fastdeploy-tmivx 目录，表示基于 PadddleLite TIM-VX 的 FastDeploy 库编译完成。
-
-RK1126 上部署 PaddleClas 分类模型请参考：[PaddleClas RK1126开发板 C++ 部署示例](../../../examples/vision/classification/paddleclas/rk1126/README.md)
--- a/docs/cn/build_and_install/rv1126.md
+++ b/docs/cn/build_and_install/rv1126.md
@@ -0,0 +1,100 @@
+# 瑞芯微 RV1126 部署环境编译安装
+
+FastDeploy基于 Paddle-Lite 后端支持在瑞芯微（Rockchip）Soc 上进行部署推理。
+更多详细的信息请参考：[PaddleLite部署示例](https://www.paddlepaddle.org.cn/lite/develop/demo_guides/verisilicon_timvx.html)。
+
+本文档介绍如何编译基于 PaddleLite 的 C++ FastDeploy 交叉编译库。
+
+相关编译选项说明如下：  
+|编译选项|默认值|说明|备注|  
+|:---|:---|:---|:---|  
+|ENABLE_LITE_BACKEND|OFF|编译RK库时需要设置为ON| - |
+|ENABLE_TIMVX|OFF|编译RK库时需要设置为ON| - |
+
+更多编译选项请参考[FastDeploy编译选项说明](./README.md)
+
+## 交叉编译环境搭建
+
+### 宿主机环境需求  
+- os：Ubuntu == 16.04
+- cmake： version >= 3.10.0  
+
+### 环境搭建
+```bash
+ # 1. Install basic software
+apt update
+apt-get install -y --no-install-recommends \
+  gcc g++ git make wget python unzip
+
+# 2. Install arm gcc toolchains
+apt-get install -y --no-install-recommends \
+  g++-arm-linux-gnueabi gcc-arm-linux-gnueabi \
+  g++-arm-linux-gnueabihf gcc-arm-linux-gnueabihf \
+  gcc-aarch64-linux-gnu g++-aarch64-linux-gnu
+
+# 3. Install cmake 3.10 or above
+wget -c https://mms-res.cdn.bcebos.com/cmake-3.10.3-Linux-x86_64.tar.gz && \
+  tar xzf cmake-3.10.3-Linux-x86_64.tar.gz && \
+  mv cmake-3.10.3-Linux-x86_64 /opt/cmake-3.10 && \
+  ln -s /opt/cmake-3.10/bin/cmake /usr/bin/cmake && \
+  ln -s /opt/cmake-3.10/bin/ccmake /usr/bin/ccmake
+```
+
+## 基于 PaddleLite 的 FastDeploy 交叉编译库编译
+搭建好交叉编译环境之后，编译命令如下：
+```bash
+# Download the latest source code
+git clone https://github.com/PaddlePaddle/FastDeploy.git
+cd FastDeploy  
+mkdir build && cd build
+
+# CMake configuration with RK toolchain
+cmake -DCMAKE_TOOLCHAIN_FILE=./../cmake/timvx.cmake \
+      -DENABLE_TIMVX=ON  \
+      -DCMAKE_INSTALL_PREFIX=fastdeploy-tmivx \
+      -DENABLE_VISION=ON \ # 是否编译集成视觉模型的部署模块，可选择开启
+      -Wno-dev ..
+
+# Build FastDeploy RV1126 C++ SDK
+make -j8
+make install
+```  
+编译完成之后，会生成 fastdeploy-tmivx 目录，表示基于 PadddleLite TIM-VX 的 FastDeploy 库编译完成。
+
+## 准备设备运行环境
+部署前要保证芯原 Linux Kernel NPU 驱动 galcore.so 版本及所适用的芯片型号与依赖库保持一致，在部署前，请登录开发板，并通过命令行输入以下命令查询 NPU 驱动版本，Rockchip建议的驱动版本为：	6.4.6.5
+```bash
+dmesg | grep Galcore
+```  
+
+如果当前版本不符合上述，请用户仔细阅读以下内容，以保证底层 NPU 驱动环境正确。
+
+有两种方式可以修改当前的 NPU 驱动版本：
+1. 手动替换 NPU 驱动版本。（推荐）
+2. 刷机，刷取 NPU 驱动版本符合要求的固件。
+
+### 手动替换 NPU 驱动版本
+1. 使用如下命令下载解压 PaddleLite demo，其中提供了现成的驱动文件
+```bash
+wget https://paddlelite-demo.bj.bcebos.com/devices/generic/PaddleLite-generic-demo.tar.gz
+tar -xf PaddleLite-generic-demo.tar.gz
+```
+2. 使用 `uname -a` 查看 `Linux Kernel` 版本，确定为 `Linux` 系统 4.19.111 版本，
+3. 将 `PaddleLite-generic-demo/libs/PaddleLite/linux/armhf/lib/verisilicon_timvx/viv_sdk_6_4_6_5/lib/1126/4.19.111/` 路径下的 `galcore.ko` 上传至开发板。
+
+4. 登录开发板，命令行输入 `sudo rmmod galcore` 来卸载原始驱动，输入 `sudo insmod galcore.ko` 来加载传上设备的驱动。（是否需要 sudo 根据开发板实际情况，部分 adb 链接的设备请提前 adb root）。此步骤如果操作失败，请跳转至方法 2。
+5. 在开发板中输入 `dmesg | grep Galcore` 查询 NPU 驱动版本，确定为：6.4.6.5
+
+### 刷机
+根据具体的开发板型号，向开发板卖家或官网客服索要 6.4.6.5 版本 NPU 驱动对应的固件和刷机方法。
+
+更多细节请参考：[PaddleLite准备设备环境](https://www.paddlepaddle.org.cn/lite/develop/demo_guides/verisilicon_timvx.html#zhunbeishebeihuanjing)
+
+## 基于 FastDeploy 在 RV1126 上的部署示例
+1. RV1126 上部署 PaddleClas 分类模型请参考：[PaddleClas 分类模型在 RV1126 上的 C++ 部署示例](../../../examples/vision/classification/paddleclas/rv1126/README.md)
+
+2. RV1126 上部署 PPYOLOE 检测模型请参考：[PPYOLOE 检测模型在 RV1126 上的 C++ 部署示例](../../../examples/vision/detection/paddledetection/rv1126/README.md)
+
+3. RV1126 上部署 YOLOv5 检测模型请参考：[YOLOv5 检测模型在 RV1126 上的 C++ 部署示例](../../../examples/vision/detection/yolov5/rv1126/README.md)
+
+4. RV1126 上部署 PP-LiteSeg 分割模型请参考：[PP-LiteSeg 分割模型在 RV1126 上的 C++ 部署示例](../../../examples/vision/segmentation/paddleseg/rv1126/README.md)
--- a/examples/vision/classification/paddleclas/quantize/README.md
+++ b/examples/vision/classification/paddleclas/quantize/README.md
@@ -4,7 +4,7 @@ FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工

 ## FastDeploy一键模型自动化压缩工具
 FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
-详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/common_tools/auto_compression/)
 注意: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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。

 ## 下载量化完成的PaddleClas模型
--- a/examples/vision/classification/paddleclas/quantize/cpp/README.md
+++ b/examples/vision/classification/paddleclas/quantize/cpp/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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.)
+- 2. 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: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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.)

 ## 以量化后的ResNet50_Vd模型为例, 进行部署，支持此模型需保证FastDeploy版本0.7.0以上(x.x.x>=0.7.0)
 在本目录执行如下命令即可完成编译,以及量化模型部署.
--- a/examples/vision/classification/paddleclas/quantize/python/README.md
+++ b/examples/vision/classification/paddleclas/quantize/python/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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.)
+- 2. 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: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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.)


 ## 以量化后的ResNet50_Vd模型为例, 进行部署
--- a/examples/vision/classification/paddleclas/rv1126/README.md
+++ b/examples/vision/classification/paddleclas/rv1126/README.md
@@ -1,11 +1,11 @@
-# PaddleClas 量化模型在 RK1126 上的部署
-目前 FastDeploy 已经支持基于 PaddleLite 部署 PaddleClas 量化模型到 RK1126 上。
+# PaddleClas 量化模型在 RV1126 上的部署
+目前 FastDeploy 已经支持基于 PaddleLite 部署 PaddleClas 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)


 ## 详细部署文档

-在 RK1126 上只支持 C++ 的部署。
+在 RV1126 上只支持 C++ 的部署。

 - [C++部署](cpp)
--- a/examples/vision/classification/paddleclas/rv1126/cpp/CMakeLists.txt
+++ b/examples/vision/classification/paddleclas/rv1126/cpp/CMakeLists.txt
--- a/examples/vision/classification/paddleclas/rv1126/cpp/README.md
+++ b/examples/vision/classification/paddleclas/rv1126/cpp/README.md
@@ -1,22 +1,22 @@
-# PaddleClas RK1126开发板 C++ 部署示例
-本目录下提供的 `infer.cc`，可以帮助用户快速完成 PaddleClas 量化模型在 RK1126 上的部署推理加速。
+# PaddleClas RV1126 开发板 C++ 部署示例
+本目录下提供的 `infer.cc`，可以帮助用户快速完成 PaddleClas 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
- 1. 软硬件环境满足要求，以及交叉编译环境的准备，请参考：[FastDeploy 交叉编译环境准备](../../../../../../docs/cn/build_and_install/rk1126.md#交叉编译环境搭建)  
+- 1. 软硬件环境满足要求，以及交叉编译环境的准备，请参考：[FastDeploy 交叉编译环境准备](../../../../../../docs/cn/build_and_install/rv1126.md#交叉编译环境搭建)  

 ### 量化模型准备
 - 1. 用户可以直接使用由 FastDeploy 提供的量化模型进行部署。
- 2. 用户可以使用 FastDeploy 提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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.)
+- 2. 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: 推理量化后的分类模型仍然需要FP32模型文件夹下的inference_cls.yaml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.)
 - 更多量化相关相关信息可查阅[模型量化](../../quantize/README.md)

-## 在 RK1126 上部署量化后的 ResNet50_Vd 分类模型
-请按照以下步骤完成在 RK1126 上部署 ResNet50_Vd 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/rk1126.md#基于-paddlelite-的-fastdeploy-交叉编译库编译)
+## 在 RV1126 上部署量化后的 ResNet50_Vd 分类模型
+请按照以下步骤完成在 RV1126 上部署 ResNet50_Vd 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：
 ```bash
-cp -r FastDeploy/build/fastdeploy-tmivx/ FastDeploy/examples/vision/classification/paddleclas/rk1126/cpp/
+cp -r FastDeploy/build/fastdeploy-tmivx/ FastDeploy/examples/vision/classification/paddleclas/rv1126/cpp/
 ```

 3. 在当前路径下载部署所需的模型和示例图片：
@@ -32,7 +32,7 @@ cp -r ILSVRC2012_val_00000010.jpeg images
 4. 编译部署示例，可使入如下命令：
 ```bash
 mkdir build && cd build
-cmake -DCMAKE_TOOLCHAIN_FILE=../fastdeploy-tmivx/timvx.cmake -DFASTDEPLOY_INSTALL_DIR=fastdeploy-tmivx ..
+cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../fastdeploy-tmivx/timvx.cmake -DFASTDEPLOY_INSTALL_DIR=${PWD}/../fastdeploy-tmivx ..
 make -j8
 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
@@ -41,7 +41,7 @@ make install
 5. 基于 adb 工具部署 ResNet50_vd 分类模型到 Rockchip RV1126，可使用如下命令：
 ```bash
 # 进入 install 目录
-cd FastDeploy/examples/vision/classification/paddleclas/rk1126/cpp/build/install/
+cd FastDeploy/examples/vision/classification/paddleclas/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 ResNet50_vd_infer ILSVRC2012_val_00000010.jpeg $DEVICE_ID
 ```
@@ -50,4 +50,4 @@ bash run_with_adb.sh infer_demo ResNet50_vd_infer ILSVRC2012_val_00000010.jpeg $

 <img width="640" src="https://user-images.githubusercontent.com/30516196/200767389-26519e50-9e4f-4fe1-8d52-260718f73476.png">

-需要特别注意的是，在 RK1126 上部署的模型需要是量化后的模型，模型的量化请参考：[模型量化](../../../../../../docs/cn/quantize.md)
+需要特别注意的是，在 RV1126 上部署的模型需要是量化后的模型，模型的量化请参考：[模型量化](../../../../../../docs/cn/quantize.md)
--- a/examples/vision/classification/paddleclas/rv1126/cpp/infer.cc
+++ b/examples/vision/classification/paddleclas/rv1126/cpp/infer.cc
--- a/examples/vision/classification/paddleclas/rv1126/cpp/run_with_adb.sh
+++ b/examples/vision/classification/paddleclas/rv1126/cpp/run_with_adb.sh
--- a/examples/vision/detection/paddledetection/quantize/README.md
+++ b/examples/vision/detection/paddledetection/quantize/README.md
@@ -4,7 +4,7 @@ FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工

 ## FastDeploy一键模型自动化压缩工具
 FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
-详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/common_tools/auto_compression/)

 ## 下载量化完成的PP-YOLOE-l模型
 用户也可以直接下载下表中的量化模型进行部署.(点击模型名字即可下载)
--- a/examples/vision/detection/paddledetection/quantize/cpp/README.md
+++ b/examples/vision/detection/paddledetection/quantize/cpp/README.md
@@ -9,7 +9,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要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.)
+- 2. 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.)

 ## 以量化后的PP-YOLOE-l模型为例, 进行部署。支持此模型需保证FastDeploy版本0.7.0以上(x.x.x>=0.7.0)
 在本目录执行如下命令即可完成编译,以及量化模型部署.
--- a/examples/vision/detection/paddledetection/quantize/python/README.md
+++ b/examples/vision/detection/paddledetection/quantize/python/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要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.)
+- 2. 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: 推理量化后的分类模型仍然需要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.)


 ## 以量化后的PP-YOLOE-l模型为例, 进行部署
--- a/examples/vision/detection/paddledetection/rk1126/download_models_and_libs.sh
+++ b/examples/vision/detection/paddledetection/rk1126/download_models_and_libs.sh
@@ -1,25 +0,0 @@
-#!/bin/bash
-
-set -e
-
-DETECTION_MODEL_DIR="$(pwd)/picodet_detection/models"
-LIBS_DIR="$(pwd)"
-
-DETECTION_MODEL_URL="https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/models/picodetv2_relu6_coco_no_fuse.tar.gz"
-LIBS_URL="https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/Paddle-Lite-libs.tar.gz"
-
-download_and_uncompress() {
-  local url="$1"
-  local dir="$2"
-  
-  echo "Start downloading ${url}"
-  curl -L ${url} > ${dir}/download.tar.gz
-  cd ${dir}
-  tar -zxvf download.tar.gz
-  rm -f download.tar.gz
-}
-
-download_and_uncompress "${DETECTION_MODEL_URL}" "${DETECTION_MODEL_DIR}"
-download_and_uncompress "${LIBS_URL}" "${LIBS_DIR}"
-
-echo "Download successful!"
--- a/examples/vision/detection/paddledetection/rk1126/picodet_detection/CMakeLists.txt
+++ b/examples/vision/detection/paddledetection/rk1126/picodet_detection/CMakeLists.txt
@@ -1,62 +0,0 @@
-cmake_minimum_required(VERSION 3.10)
-set(CMAKE_SYSTEM_NAME Linux)
-if(TARGET_ARCH_ABI STREQUAL "armv8")
-    set(CMAKE_SYSTEM_PROCESSOR aarch64)
-    set(CMAKE_C_COMPILER "aarch64-linux-gnu-gcc")
-    set(CMAKE_CXX_COMPILER "aarch64-linux-gnu-g++")
-elseif(TARGET_ARCH_ABI STREQUAL "armv7hf")
-    set(CMAKE_SYSTEM_PROCESSOR arm)
-    set(CMAKE_C_COMPILER "arm-linux-gnueabihf-gcc")
-    set(CMAKE_CXX_COMPILER "arm-linux-gnueabihf-g++")
-else()
-    message(FATAL_ERROR "Unknown arch abi ${TARGET_ARCH_ABI}, only support armv8 and armv7hf.")
-    return()
-endif()
-
-project(object_detection_demo)
-
-message(STATUS "TARGET ARCH ABI: ${TARGET_ARCH_ABI}")
-message(STATUS "PADDLE LITE DIR: ${PADDLE_LITE_DIR}")
-include_directories(${PADDLE_LITE_DIR}/include)
-link_directories(${PADDLE_LITE_DIR}/libs/${TARGET_ARCH_ABI})
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
-if(TARGET_ARCH_ABI STREQUAL "armv8")
-    set(CMAKE_CXX_FLAGS "-march=armv8-a ${CMAKE_CXX_FLAGS}")
-    set(CMAKE_C_FLAGS "-march=armv8-a ${CMAKE_C_FLAGS}")
-elseif(TARGET_ARCH_ABI STREQUAL "armv7hf")
-    set(CMAKE_CXX_FLAGS "-march=armv7-a -mfloat-abi=hard -mfpu=neon-vfpv4 ${CMAKE_CXX_FLAGS}")
-    set(CMAKE_C_FLAGS "-march=armv7-a -mfloat-abi=hard -mfpu=neon-vfpv4 ${CMAKE_C_FLAGS}" )
-endif()
-
-include_directories(${PADDLE_LITE_DIR}/libs/${TARGET_ARCH_ABI}/third_party/yaml-cpp/include)
-link_directories(${PADDLE_LITE_DIR}/libs/${TARGET_ARCH_ABI}/third_party/yaml-cpp)
-
-find_package(OpenMP REQUIRED)
-if(OpenMP_FOUND OR OpenMP_CXX_FOUND)
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
-    message(STATUS "Found OpenMP ${OpenMP_VERSION} ${OpenMP_CXX_VERSION}")
-    message(STATUS "OpenMP C flags:  ${OpenMP_C_FLAGS}")
-    message(STATUS "OpenMP CXX flags:  ${OpenMP_CXX_FLAGS}")
-    message(STATUS "OpenMP OpenMP_CXX_LIB_NAMES:  ${OpenMP_CXX_LIB_NAMES}")
-    message(STATUS "OpenMP OpenMP_CXX_LIBRARIES:  ${OpenMP_CXX_LIBRARIES}")
-else()
-    message(FATAL_ERROR "Could not found OpenMP!")
-    return()
-endif()
-find_package(OpenCV REQUIRED)
-if(OpenCV_FOUND OR OpenCV_CXX_FOUND)
-    include_directories(${OpenCV_INCLUDE_DIRS})
-    message(STATUS "OpenCV library status:")
-    message(STATUS "    version: ${OpenCV_VERSION}")
-    message(STATUS "    libraries: ${OpenCV_LIBS}")
-    message(STATUS "    include path: ${OpenCV_INCLUDE_DIRS}")
-else()
-    message(FATAL_ERROR "Could not found OpenCV!")
-    return()
-endif()
-
-
-add_executable(object_detection_demo object_detection_demo.cc)
-
-target_link_libraries(object_detection_demo paddle_full_api_shared dl ${OpenCV_LIBS} yaml-cpp)
--- a/examples/vision/detection/paddledetection/rk1126/picodet_detection/README.md
+++ b/examples/vision/detection/paddledetection/rk1126/picodet_detection/README.md
@@ -1,343 +0,0 @@
-# 目标检测 C++ API Demo 使用指南
-
-在 ARMLinux 上实现实时的目标检测功能，此 Demo 有较好的的易用性和扩展性，如在 Demo 中跑自己训练好的模型等。
- - 如果该开发板使用搭载了芯原 NPU （瑞芯微、晶晨、JQL、恩智浦）的 Soc，将有更好的加速效果。
-
-## 如何运行目标检测 Demo
-
-### 环境准备
-
-* 准备 ARMLiunx 开发版，将系统刷为 Ubuntu，用于 Demo 编译和运行。请注意，本 Demo 是使用板上编译，而非交叉编译，因此需要图形界面的开发板操作系统。
-* 如果需要使用 芯原 NPU 的计算加速，对 NPU 驱动版本有严格要求，请务必注意事先参考 [芯原 TIM-VX 部署示例](https://paddle-lite.readthedocs.io/zh/develop/demo_guides/verisilicon_timvx.html#id6)，将 NPU 驱动改为要求的版本。
-* Paddle Lite 当前已验证的开发板为 Khadas VIM3（芯片为 Amlogic A311d）、荣品 RV1126、荣品RV1109，其它平台用户可自行尝试；
- - Khadas VIM3：由于 VIM3 出厂自带 Android 系统，请先刷成 Ubuntu 系统，在此提供刷机教程：[VIM3/3L Linux 文档](https://docs.khadas.com/linux/zh-cn/vim3)，其中有详细描述刷机方法。以及系统镜像：VIM3 Linux：VIM3_Ubuntu-gnome-focal_Linux-4.9_arm64_EMMC_V1.0.7-210625：[官方链接](http://dl.khadas.com/firmware/VIM3/Ubuntu/EMMC/VIM3_Ubuntu-gnome-focal_Linux-4.9_arm64_EMMC_V1.0.7-210625.img.xz)；[百度云备用链接](https://paddlelite-demo.bj.bcebos.com/devices/verisilicon/firmware/khadas/vim3/VIM3_Ubuntu-gnome-focal_Linux-4.9_arm64_EMMC_V1.0.7-210625.img.xz)
- - 荣品 RV1126、1109：由于出场自带 buildroot 系统，如果使用 GUI 界面的 demo，请先刷成 Ubuntu 系统，在此提供刷机教程：[RV1126/1109 教程](https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/os_img/rockchip/RV1126-RV1109%E4%BD%BF%E7%94%A8%E6%8C%87%E5%AF%BC%E6%96%87%E6%A1%A3-V3.0.pdf)，[刷机工具](https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/os_img/rockchip/RKDevTool_Release.zip)，以及镜像：[1126镜像](https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/os_img/update-pro-rv1126-ubuntu20.04-5-720-1280-v2-20220505.img)，[1109镜像](https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/os_img/update-pro-rv1109-ubuntu20.04-5.5-720-1280-v2-20220429.img)。完整的文档和各种镜像请参考[百度网盘链接](https://pan.baidu.com/s/1Id0LMC0oO2PwR2YcYUAaiQ#list/path=%2F&parentPath=%2Fsharelink2521613171-184070898837664)，密码：2345。
-* 准备 usb camera，注意使用 openCV capture 图像时，请注意 usb camera 的 video序列号作为入参。
-* 请注意，瑞芯微芯片不带有 HDMI 接口，图像显示是依赖 MIPI DSI，所以请准备好 MIPI 显示屏（我们提供的镜像是 720*1280 分辨率，网盘中有更多分辨率选择，注意：请选择 camera-gc2093x2 的镜像）。
-* 配置开发板的网络。如果是办公网络红区，可以将开发板和PC用以太网链接，然后PC共享网络给开发板。
-* gcc g++ opencv cmake 的安装（以下所有命令均在设备上操作）
-
-```bash
-$ sudo apt-get update
-$ sudo apt-get install gcc g++ make wget unzip libopencv-dev pkg-config
-$ wget https://www.cmake.org/files/v3.10/cmake-3.10.3.tar.gz
-$ tar -zxvf cmake-3.10.3.tar.gz
-$ cd cmake-3.10.3
-$ ./configure
-$ make
-$ sudo make install
-```
-
-### 部署步骤
-
-1. 将本 repo 上传至 VIM3 开发板，或者直接开发板上下载或者 git clone 本 repo
-2. 目标检测 Demo 位于 `Paddle-Lite-Demo/object_detection/linux/picodet_detection` 目录
-3. 进入 `Paddle-Lite-Demo/object_detection/linux` 目录, 终端中执行 `download_models_and_libs.sh` 脚本自动下载模型和 Paddle Lite 预测库
-
-```shell
-cd Paddle-Lite-Demo/object_detection/linux   # 1. 终端中进入 Paddle-Lite-Demo/object_detection/linux
-sh download_models_and_libs.sh               # 2. 执行脚本下载依赖项 （需要联网）
-```
-
-下载完成后会出现提示： `Download successful!`
-4. 执行用例(保证 ARMLinux 环境准备完成)
-
-```shell
-cd picodet_detection        # 1. 终端中进入
-sh build.sh armv8           # 2. 编译 Demo 可执行程序，默认编译 armv8，如果是 32bit 环境，则改成 sh build.sh armv7hf。
-sh run.sh armv8             # 3. 执行物体检测（picodet 模型） demo，会直接开启摄像头，启动图形界面并呈现检测结果。如果是 32bit 环境，则改成 sh run.sh armv7hf
-```
-
-### Demo 结果如下:（注意，示例的 picodet 仅使用 coco 数据集，在实际场景中效果一般，请使用实际业务场景重新训练）
-
-  <img src="https://paddlelite-demo.bj.bcebos.com/Paddle-Lite-Demo/demo_view.jpg" alt="demo_view" style="zoom: 10%;" />
-
-## 更新预测库
-
-* Paddle Lite 项目：https://github.com/PaddlePaddle/Paddle-Lite
- * 参考 [芯原 TIM-VX 部署示例](https://paddle-lite.readthedocs.io/zh/develop/demo_guides/verisilicon_timvx.html#tim-vx)，编译预测库
- * 编译最终产物位于 `build.lite.xxx.xxx.xxx` 下的 `inference_lite_lib.xxx.xxx`
-    * 替换 c++ 库
-        * 头文件
-          将生成的 `build.lite.linux.armv8.gcc/inference_lite_lib.armlinux.armv8.nnadapter/cxx/include` 文件夹替换 Demo 中的 `Paddle-Lite-Demo/object_detection/linux/Paddle-Lite/include`
-        * armv8
-          将生成的 `build.lite.linux.armv8.gcc/inference_lite_lib.armlinux.armv8.nnadapter/cxx/libs/libpaddle_full_api_shared.so、libnnadapter.so、libtim-vx.so、libverisilicon_timvx.so` 库替换 Demo 中的 `Paddle-Lite-Demo/object_detection/linux/Paddle-Lite/libs/armv8/` 目录下同名 so
-        * armv7hf
-          将生成的 `build.lite.linux.armv7hf.gcc/inference_lite_lib.armlinux.armv7hf.nnadapter/cxx/libs/libpaddle_full_api_shared.so、libnnadapter.so、libtim-vx.so、libverisilicon_timvx.so` 库替换 Demo 中的 `Paddle-Lite-Demo/object_detection/linux/Paddle-Lite/libs/armv7hf/` 目录下同名 so
-
-## Demo 内容介绍
-
-先整体介绍下目标检测 Demo 的代码结构，然后再简要地介绍 Demo 每部分功能.
-
-1. `object_detection_demo.cc`： C++ 预测代码
-
-```shell
-# 位置：
-Paddle-Lite-Demo/object_detection/linux/picodet_detection/object_detection_demo.cc
-```
-
-2. `models` : 模型文件夹 (执行 download_models_and_libs.sh 后会下载 picodet Paddle 模型), label 使用 Paddle-Lite-Demo/object_detection/assets/labels 目录下 coco_label_list.txt
-
-```shell
-# 位置：
-Paddle-Lite-Demo/object_detection/linux/picodet_detection/models/picodetv2_relu6_coco_no_fuse
-Paddle-Lite-Demo/object_detection/assets/labels/coco_label_list.txt
-```
-
-3. `Paddle-Lite`：内含 Paddle-Lite 头文件和 动态库，默认带有 timvx 加速库，以及第三方库 yaml-cpp 用于解析 yml 配置文件（执行 download_models_and_libs.sh 后会下载）
-
-```shell
-# 位置
-# 如果要替换动态库 so，则将新的动态库 so 更新到此目录下
-Paddle-Lite-Demo/object_detection/linux/Paddle-Lite/libs/armv8
-Paddle-Lite-Demo/object_detection/linux/Paddle-Lite/include
-```
-
-4. `CMakeLists.txt` : C++ 预测代码的编译脚本，用于生成可执行文件
-
-```shell
-# 位置
-Paddle-Lite-Demo/object_detection/linux/picodet_detection/CMakeLists.txt
-# 如果有cmake 编译选项更新，可以在 CMakeLists.txt 进行修改即可，默认编译 armv8 可执行文件；
-```
-
-5. `build.sh` : 编译脚本
-
-```shell
-# 位置
-Paddle-Lite-Demo/object_detection/linux/picodet_detection/build.sh
-```
-
-6. `run.sh` : 运行脚本，请注意设置 arm-aarch，armv8 或者 armv7hf。默认为armv8
-
-```shell
-# 位置
-Paddle-Lite-Demo/object_detection/linux/picodet_detection/run.sh
-```
- 请注意，运行需要5个元素：测试程序、模型、label 文件、异构配置、yaml 文件。
-
-## 代码讲解 （使用 Paddle Lite `C++ API` 执行预测）
-
-ARMLinux 示例基于 C++ API 开发，调用 Paddle Lite `C++s API` 包括以下五步。更详细的 `API` 描述参考：[Paddle Lite C++ API ](https://paddle-lite.readthedocs.io/zh/latest/api_reference/cxx_api_doc.html)。
-
-```c++
-#include <iostream>
-// 引入 C++ API
-#include "include/paddle_api.h"
-#include "include/paddle_use_ops.h"
-#include "include/paddle_use_kernels.h"
-
-// 使用在线编译模型的方式（等价于使用 opt 工具）
-
-// 1. 设置 CxxConfig
-paddle::lite_api::CxxConfig cxx_config;
-std::vector<paddle::lite_api::Place> valid_places;
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kARM), PRECISION(kInt8)});
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kARM), PRECISION(kFloat)});
-// 如果只需要 cpu 计算，那到此结束即可，下面是设置 NPU 的代码段
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kNNAdapter), PRECISION(kInt8)});
-cxx_config.set_valid_places(valid_places);
-std::string device = "verisilicon_timvx";
-cxx_config.set_nnadapter_device_names({device});
-// 设置定制化的异构策略 （如需要）
-cxx_config.set_nnadapter_subgraph_partition_config_buffer(
-            nnadapter_subgraph_partition_config_string);
-
-// 2. 生成 nb 模型 （等价于 opt 工具的产出）
-std::shared_ptr<paddle::lite_api::PaddlePredictor> predictor = nullptr;
-predictor = paddle::lite_api::CreatePaddlePredictor(cxx_config);
-predictor->SaveOptimizedModel(
-        model_path, paddle::lite_api::LiteModelType::kNaiveBuffer);
-
-// 3. 设置 MobileConfig
-MobileConfig config;
-config.set_model_from_file(modelPath); // 设置 NaiveBuffer 格式模型路径
-config.set_power_mode(LITE_POWER_NO_BIND); // 设置 CPU 运行模式
-config.set_threads(4); // 设置工作线程数
-
-// 4. 创建 PaddlePredictor
-predictor = CreatePaddlePredictor<MobileConfig>(config);
-
-// 5. 设置输入数据，注意，如果是带后处理的 picodet ，则是有两个输入
-std::unique_ptr<Tensor> input_tensor(std::move(predictor->GetInput(0)));
-input_tensor->Resize({1, 3, 416, 416});
-auto* data = input_tensor->mutable_data<float>();
-// scale_factor tensor
-auto scale_factor_tensor = predictor->GetInput(1);
-scale_factor_tensor->Resize({1, 2});
-auto scale_factor_data = scale_factor_tensor->mutable_data<float>();
-scale_factor_data[0] = 1.0f;
-scale_factor_data[1] = 1.0f;
-
-// 6. 执行预测
-predictor->run();
-
-// 7. 获取输出数据
-std::unique_ptr<const Tensor> output_tensor(std::move(predictor->GetOutput(0)));
-
-```
-
-## 如何更新模型和输入/输出预处理
-
-### 更新模型
-1. 请参考 PaddleDetection 中 [picodet 重训和全量化文档](https://github.com/PaddlePaddle/PaddleDetection/blob/develop/configs/picodet/FULL_QUANTIZATION.md)，基于用户自己数据集重训并且重新全量化
-2. 将模型存放到目录 `object_detection_demo/models/` 下；
-3. 模型名字跟工程中模型名字一模一样，即均是使用 `model`、`params`；
-
-```shell
-# shell 脚本 `object_detection_demo/run.sh`
-TARGET_ABI=armv8 # for 64bit, such as Amlogic A311D
-#TARGET_ABI=armv7hf # for 32bit, such as Rockchip 1109/1126
-if [ -n "$1" ]; then
-    TARGET_ABI=$1
-fi
-export LD_LIBRARY_PATH=../Paddle-Lite/libs/$TARGET_ABI/
-export GLOG_v=0 # Paddle-Lite 日志等级
-export VSI_NN_LOG_LEVEL=0 # TIM-VX 日志等级
-export VIV_VX_ENABLE_GRAPH_TRANSFORM=-pcq:1 # NPU 开启 perchannel 量化模型
-export VIV_VX_SET_PER_CHANNEL_ENTROPY=100 # 同上
-build/object_detection_demo models/picodetv2_relu6_coco_no_fuse ../../assets/labels/coco_label_list.txt models/picodetv2_relu6_coco_no_fuse/subgraph.txt models/picodetv2_relu6_coco_no_fuse/picodet.yml  # 执行 Demo 程序，4个 arg 分别为：模型、 label 文件、 自定义异构配置、 yaml
-```
-
- 如果需要更新 `label_list` 或者 `yaml` 文件，则修改 `object_detection_demo/run.sh` 中执行命令的第二个和第四个 arg 指定为新的 label 文件和 yaml 配置文件；
-
-```shell
-# 代码文件 `object_detection_demo/rush.sh`
-export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${PADDLE_LITE_DIR}/libs/${TARGET_ARCH_ABI}
-build/object_detection_demo {模型} {label} {自定义异构配置文件} {yaml}
-```
-
-### 更新输入/输出预处理
-
-1. 更新输入预处理
-预处理完全根据 yaml 文件来，如果完全按照 PaddleDetection 中 picodet 重训，只需要替换 yaml 文件即可
-
-2. 更新输出预处理
-此处需要更新 `object_detection_demo/object_detection_demo.cc` 中的 `postprocess` 方法
-
-```c++
-std::vector<RESULT> postprocess(const float *output_data, int64_t output_size,
-                                const std::vector<std::string> &word_labels,
-                                const float score_threshold,
-                                cv::Mat &output_image, double time) {
-  std::vector<RESULT> results;
-  std::vector<cv::Scalar> colors = {
-      cv::Scalar(237, 189, 101), cv::Scalar(0, 0, 255),
-      cv::Scalar(102, 153, 153), cv::Scalar(255, 0, 0),
-      cv::Scalar(9, 255, 0),     cv::Scalar(0, 0, 0),
-      cv::Scalar(51, 153, 51)};
-  for (int64_t i = 0; i < output_size; i += 6) {
-    if (output_data[i + 1] < score_threshold) {
-      continue;
-    }
-    int class_id = static_cast<int>(output_data[i]);
-    float score = output_data[i + 1];
-    RESULT result;
-    std::string class_name = "Unknown";
-    if (word_labels.size() > 0 && class_id >= 0 &&
-        class_id < word_labels.size()) {
-      class_name = word_labels[class_id];
-    }
-    result.class_name = class_name;
-    result.score = score;
-    result.left = output_data[i + 2] / 416; // 此处416根据输入的 HW 得来
-    result.top = output_data[i + 3] / 416;
-    result.right = output_data[i + 4] / 416;
-    result.bottom = output_data[i + 5] / 416;
-    int lx = static_cast<int>(result.left * output_image.cols);
-    int ly = static_cast<int>(result.top * output_image.rows);
-    int w = static_cast<int>(result.right * output_image.cols) - lx;
-    int h = static_cast<int>(result.bottom * output_image.rows) - ly;
-    cv::Rect bounding_box =
-        cv::Rect(lx, ly, w, h) &
-        cv::Rect(0, 0, output_image.cols, output_image.rows);
-    if (w > 0 && h > 0 && score <= 1) {
-      cv::Scalar color = colors[results.size() % colors.size()];
-      cv::rectangle(output_image, bounding_box, color);
-      cv::rectangle(output_image, cv::Point2d(lx, ly),
-                    cv::Point2d(lx + w, ly - 10), color, -1);
-      cv::putText(output_image, std::to_string(results.size()) + "." +
-                                    class_name + ":" + std::to_string(score),
-                  cv::Point2d(lx, ly), cv::FONT_HERSHEY_PLAIN, 1,
-                  cv::Scalar(255, 255, 255));
-      results.push_back(result);
-    }
-  }
-  return results;
-}
-```
-
-## 更新模型后，自定义 NPU-CPU 异构配置（如需使用 NPU 加速）
-由于使用芯原 NPU 在 8bit 量化的情况下有最优的性能，因此部署时，我们往往会考虑量化
- - 由于量化可能会引入一定程度的精度问题，所以我们可以通过自定义的异构定制，来将部分有精度问题的 layer 异构至cpu，从而达到最优的精度
-
-### 第一步，确定模型量化后在 arm cpu 上的精度
-如果在 arm cpu 上，精度都无法满足，那量化本身就是失败的，此时可以考虑修改训练集或者预处理。
- - 修改 Demo 程序，仅用 arm cpu 计算
-```c++
-paddle::lite_api::CxxConfig cxx_config;
-std::vector<paddle::lite_api::Place> valid_places;
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kARM), PRECISION(kInt8)});
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kARM), PRECISION(kFloat)});
-// 仅用 arm cpu 计算， 注释如下代码即可
-/*
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kNNAdapter), PRECISION(kInt8)});
-valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kNNAdapter), PRECISION(kFloat)});
-*/
-```
-如果 arm cpu 计算结果精度达标，则继续
-
-### 第二步，获取整网拓扑信息
- - 回退第一步的修改，不再注释，使用 NPU 加速
- - 运行 Demo，如果此时精度良好，则无需参考后面步骤，模型部署和替换非常顺利，enjoy it。
- - 如果精度不行，请参考后续步骤。
-
-### 第三步，获取整网拓扑信息
- - 回退第一步的修改，使用
- - 修改 run.sh ，将其中 export GLOG_v=0 改为 export GLOG_v=5
- - 运行 Demo，等摄像头启动，即可 ctrl+c 关闭 Demo
- - 收集日志，搜索关键字 "subgraph operators" 随后那一段，便是整个模型的拓扑信息，其格式如下：
-    - 每行记录由『算子类型:输入张量名列表:输出张量名列表』组成（即以分号分隔算子类型、输入和输出张量名列表），以逗号分隔输入、输出张量名列表中的每个张量名；
-    - 示例说明：
-    ```
-      op_type0:var_name0,var_name1:var_name2          表示将算子类型为 op_type0、输入张量为var_name0 和 var_name1、输出张量为 var_name2 的节点强制运行在 ARM CPU 上
-    ```
-
-### 第四步，修改异构配置文件
- - 首先看到示例 Demo 中 Paddle-Lite-Demo/object_detection/linux/picodet_detection/models/picodetv2_relu6_coco_no_fuse 目录下的 subgraph.txt 文件。(feed 和 fetch 分别代表整个模型的输入和输入)
-  ```
-  feed:feed:scale_factor
-  feed:feed:image
-
-  sqrt:tmp_3:sqrt_0.tmp_0
-  reshape2:sqrt_0.tmp_0:reshape2_0.tmp_0,reshape2_0.tmp_1
-
-  matmul_v2:softmax_0.tmp_0,auto_113_:linear_0.tmp_0
-  reshape2:linear_0.tmp_0:reshape2_2.tmp_0,reshape2_2.tmp_1
-
-  sqrt:tmp_6:sqrt_1.tmp_0
-  reshape2:sqrt_1.tmp_0:reshape2_3.tmp_0,reshape2_3.tmp_1
-
-  matmul_v2:softmax_1.tmp_0,auto_113_:linear_1.tmp_0
-  reshape2:linear_1.tmp_0:reshape2_5.tmp_0,reshape2_5.tmp_1
-
-  sqrt:tmp_9:sqrt_2.tmp_0
-  reshape2:sqrt_2.tmp_0:reshape2_6.tmp_0,reshape2_6.tmp_1
-
-  matmul_v2:softmax_2.tmp_0,auto_113_:linear_2.tmp_0
-  ...
-  ```
- - 在 txt 中的都是需要异构至 cpu 计算的 layer，在示例 Demo 中，我们把 picodet 后处理的部分异构至 arm cpu 做计算，不必担心，Paddle-Lite 的 arm kernel 性能也是非常卓越。
- - 如果新训练的模型没有额外修改 layer，则直接复制使用示例 Demo 中的 subgraph.txt 即可
- - 此时 ./run.sh 看看精度是否符合预期，如果精度符合预期，恭喜，可以跳过本章节，enjoy it。
- - 如果精度不符合预期，则将上文『第二步，获取整网拓扑信息』中获取的拓扑信息，从 "feed" 之后第一行，直到 "sqrt" 之前，都复制进 sugraph.txt。这一步代表了将大量的 backbone 部分算子放到 arm cpu 计算。
- - 此时 ./run.sh 看看精度是否符合预期，如果精度达标，那说明在 backbone 中确实存在引入 NPU 精度异常的层（再次重申，在 subgraph.txt 的代表强制在 arm cpu 计算）。
- - 逐行删除、成片删除、二分法，发挥开发人员的耐心，找到引入 NPU 精度异常的 layer，将其留在 subgraph.txt 中，按照经验，如果有 NPU 精度问题，可能会有 1~5 层conv layer 需要异构。
- - 剩余没有精度问题的 layer 在 subgraph.txt 中删除即可
--- a/examples/vision/detection/paddledetection/rk1126/picodet_detection/build.sh
+++ b/examples/vision/detection/paddledetection/rk1126/picodet_detection/build.sh
@@ -1,17 +0,0 @@
-#!/bin/bash
-USE_FULL_API=TRUE
-# configure
-TARGET_ARCH_ABI=armv8 # for RK3399, set to default arch abi
-#TARGET_ARCH_ABI=armv7hf # for Raspberry Pi 3B
-PADDLE_LITE_DIR=../Paddle-Lite
-THIRD_PARTY_DIR=./third_party
-if [ "x$1" != "x" ]; then
-    TARGET_ARCH_ABI=$1
-fi
-
-# build
-rm -rf build
-mkdir build
-cd build
-cmake -DPADDLE_LITE_DIR=${PADDLE_LITE_DIR} -DTARGET_ARCH_ABI=${TARGET_ARCH_ABI} -DTHIRD_PARTY_DIR=${THIRD_PARTY_DIR} ..
-make
--- a/examples/vision/detection/paddledetection/rk1126/picodet_detection/object_detection_demo.cc
+++ b/examples/vision/detection/paddledetection/rk1126/picodet_detection/object_detection_demo.cc
@@ -1,411 +0,0 @@
-// Copyright (c) 2019 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 "paddle_api.h"
-#include "yaml-cpp/yaml.h"
-#include <arm_neon.h>
-#include <fstream>
-#include <limits>
-#include <opencv2/core.hpp>
-#include <opencv2/highgui.hpp>
-#include <opencv2/opencv.hpp>
-#include <stdio.h>
-#include <sys/time.h>
-#include <unistd.h>
-#include <vector>
-
-int WARMUP_COUNT = 0;
-int REPEAT_COUNT = 1;
-const int CPU_THREAD_NUM = 2;
-const paddle::lite_api::PowerMode CPU_POWER_MODE =
-    paddle::lite_api::PowerMode::LITE_POWER_HIGH;
-const std::vector<int64_t> INPUT_SHAPE = {1, 3, 416, 416};
-std::vector<float> INPUT_MEAN = {0.f, 0.f, 0.f};
-std::vector<float> INPUT_STD = {1.f, 1.f, 1.f};
-float INPUT_SCALE = 1 / 255.f;
-const float SCORE_THRESHOLD = 0.35f;
-
-struct RESULT {
-  std::string class_name;
-  float score;
-  float left;
-  float top;
-  float right;
-  float bottom;
-};
-
-inline int64_t get_current_us() {
-  struct timeval time;
-  gettimeofday(&time, NULL);
-  return 1000000LL * (int64_t)time.tv_sec + (int64_t)time.tv_usec;
-}
-
-bool read_file(const std::string &filename, std::vector<char> *contents,
-               bool binary = true) {
-  FILE *fp = fopen(filename.c_str(), binary ? "rb" : "r");
-  if (!fp)
-    return false;
-  fseek(fp, 0, SEEK_END);
-  size_t size = ftell(fp);
-  fseek(fp, 0, SEEK_SET);
-  contents->clear();
-  contents->resize(size);
-  size_t offset = 0;
-  char *ptr = reinterpret_cast<char *>(&(contents->at(0)));
-  while (offset < size) {
-    size_t already_read = fread(ptr, 1, size - offset, fp);
-    offset += already_read;
-    ptr += already_read;
-  }
-  fclose(fp);
-  return true;
-}
-
-std::vector<std::string> load_labels(const std::string &path) {
-  std::ifstream file;
-  std::vector<std::string> labels;
-  file.open(path);
-  while (file) {
-    std::string line;
-    std::getline(file, line);
-    labels.push_back(line);
-  }
-  file.clear();
-  file.close();
-  return labels;
-}
-
-bool load_yaml_config(std::string yaml_path) {
-  YAML::Node cfg;
-  try {
-    std::cout << "before loadFile" << std::endl;
-    cfg = YAML::LoadFile(yaml_path);
-  } catch (YAML::BadFile &e) {
-    std::cout << "Failed to load yaml file " << yaml_path
-              << ", maybe you should check this file." << std::endl;
-    return false;
-  }
-  auto preprocess_cfg = cfg["TestReader"]["sample_transforms"];
-  for (const auto &op : preprocess_cfg) {
-    if (!op.IsMap()) {
-      std::cout << "Require the transform information in yaml be Map type."
-                << std::endl;
-      std::abort();
-    }
-    auto op_name = op.begin()->first.as<std::string>();
-    if (op_name == "NormalizeImage") {
-      INPUT_MEAN = op.begin()->second["mean"].as<std::vector<float>>();
-      INPUT_STD = op.begin()->second["std"].as<std::vector<float>>();
-      INPUT_SCALE = op.begin()->second["scale"].as<float>();
-    }
-  }
-  return true;
-}
-
-void preprocess(cv::Mat &input_image, std::vector<float> &input_mean,
-                std::vector<float> &input_std, float input_scale,
-                int input_width, int input_height, float *input_data) {
-  cv::Mat resize_image;
-  cv::resize(input_image, resize_image, cv::Size(input_width, input_height), 0,
-             0);
-  if (resize_image.channels() == 4) {
-    cv::cvtColor(resize_image, resize_image, cv::COLOR_BGRA2RGB);
-  }
-  cv::Mat norm_image;
-  resize_image.convertTo(norm_image, CV_32FC3, input_scale);
-  // NHWC->NCHW
-  int image_size = input_height * input_width;
-  const float *image_data = reinterpret_cast<const float *>(norm_image.data);
-  float32x4_t vmean0 = vdupq_n_f32(input_mean[0]);
-  float32x4_t vmean1 = vdupq_n_f32(input_mean[1]);
-  float32x4_t vmean2 = vdupq_n_f32(input_mean[2]);
-  float32x4_t vscale0 = vdupq_n_f32(1.0f / input_std[0]);
-  float32x4_t vscale1 = vdupq_n_f32(1.0f / input_std[1]);
-  float32x4_t vscale2 = vdupq_n_f32(1.0f / input_std[2]);
-  float *input_data_c0 = input_data;
-  float *input_data_c1 = input_data + image_size;
-  float *input_data_c2 = input_data + image_size * 2;
-  int i = 0;
-  for (; i < image_size - 3; i += 4) {
-    float32x4x3_t vin3 = vld3q_f32(image_data);
-    float32x4_t vsub0 = vsubq_f32(vin3.val[0], vmean0);
-    float32x4_t vsub1 = vsubq_f32(vin3.val[1], vmean1);
-    float32x4_t vsub2 = vsubq_f32(vin3.val[2], vmean2);
-    float32x4_t vs0 = vmulq_f32(vsub0, vscale0);
-    float32x4_t vs1 = vmulq_f32(vsub1, vscale1);
-    float32x4_t vs2 = vmulq_f32(vsub2, vscale2);
-    vst1q_f32(input_data_c0, vs0);
-    vst1q_f32(input_data_c1, vs1);
-    vst1q_f32(input_data_c2, vs2);
-    image_data += 12;
-    input_data_c0 += 4;
-    input_data_c1 += 4;
-    input_data_c2 += 4;
-  }
-  for (; i < image_size; i++) {
-    *(input_data_c0++) = (*(image_data++) - input_mean[0]) / input_std[0];
-    *(input_data_c1++) = (*(image_data++) - input_mean[1]) / input_std[1];
-    *(input_data_c2++) = (*(image_data++) - input_mean[2]) / input_std[2];
-  }
-}
-
-std::vector<RESULT> postprocess(const float *output_data, int64_t output_size,
-                                const std::vector<std::string> &word_labels,
-                                const float score_threshold,
-                                cv::Mat &output_image, double time) {
-  std::vector<RESULT> results;
-  std::vector<cv::Scalar> colors = {
-      cv::Scalar(237, 189, 101), cv::Scalar(0, 0, 255),
-      cv::Scalar(102, 153, 153), cv::Scalar(255, 0, 0),
-      cv::Scalar(9, 255, 0),     cv::Scalar(0, 0, 0),
-      cv::Scalar(51, 153, 51)};
-  for (int64_t i = 0; i < output_size; i += 6) {
-    if (output_data[i + 1] < score_threshold) {
-      continue;
-    }
-    int class_id = static_cast<int>(output_data[i]);
-    float score = output_data[i + 1];
-    RESULT result;
-    std::string class_name = "Unknown";
-    if (word_labels.size() > 0 && class_id >= 0 &&
-        class_id < word_labels.size()) {
-      class_name = word_labels[class_id];
-    }
-    result.class_name = class_name;
-    result.score = score;
-    result.left = output_data[i + 2] / 416;
-    result.top = output_data[i + 3] / 416;
-    result.right = output_data[i + 4] / 416;
-    result.bottom = output_data[i + 5] / 416;
-    int lx = static_cast<int>(result.left * output_image.cols);
-    int ly = static_cast<int>(result.top * output_image.rows);
-    int w = static_cast<int>(result.right * output_image.cols) - lx;
-    int h = static_cast<int>(result.bottom * output_image.rows) - ly;
-    cv::Rect bounding_box =
-        cv::Rect(lx, ly, w, h) &
-        cv::Rect(0, 0, output_image.cols, output_image.rows);
-    if (w > 0 && h > 0 && score <= 1) {
-      cv::Scalar color = colors[results.size() % colors.size()];
-      cv::rectangle(output_image, bounding_box, color);
-      cv::rectangle(output_image, cv::Point2d(lx, ly),
-                    cv::Point2d(lx + w, ly - 10), color, -1);
-      cv::putText(output_image, std::to_string(results.size()) + "." +
-                                    class_name + ":" + std::to_string(score),
-                  cv::Point2d(lx, ly), cv::FONT_HERSHEY_PLAIN, 1,
-                  cv::Scalar(255, 255, 255));
-      results.push_back(result);
-    }
-  }
-  return results;
-}
-
-cv::Mat process(cv::Mat &input_image, std::vector<std::string> &word_labels,
-                std::shared_ptr<paddle::lite_api::PaddlePredictor> &predictor) {
-  // Preprocess image and fill the data of input tensor
-  std::unique_ptr<paddle::lite_api::Tensor> input_tensor(
-      std::move(predictor->GetInput(0)));
-  input_tensor->Resize(INPUT_SHAPE);
-  int input_width = INPUT_SHAPE[3];
-  int input_height = INPUT_SHAPE[2];
-  auto *input_data = input_tensor->mutable_data<float>();
-#if 1
-  // scale_factor tensor
-  auto scale_factor_tensor = predictor->GetInput(1);
-  scale_factor_tensor->Resize({1, 2});
-  auto scale_factor_data = scale_factor_tensor->mutable_data<float>();
-  scale_factor_data[0] = 1.0f;
-  scale_factor_data[1] = 1.0f;
-#endif
-
-  double preprocess_start_time = get_current_us();
-  preprocess(input_image, INPUT_MEAN, INPUT_STD, INPUT_SCALE, input_width,
-             input_height, input_data);
-  double preprocess_end_time = get_current_us();
-  double preprocess_time =
-      (preprocess_end_time - preprocess_start_time) / 1000.0f;
-
-  double prediction_time;
-  // Run predictor
-  // warm up to skip the first inference and get more stable time, remove it in
-  // actual products
-  for (int i = 0; i < WARMUP_COUNT; i++) {
-    predictor->Run();
-  }
-  // repeat to obtain the average time, set REPEAT_COUNT=1 in actual products
-  double max_time_cost = 0.0f;
-  double min_time_cost = std::numeric_limits<float>::max();
-  double total_time_cost = 0.0f;
-  for (int i = 0; i < REPEAT_COUNT; i++) {
-    auto start = get_current_us();
-    predictor->Run();
-    auto end = get_current_us();
-    double cur_time_cost = (end - start) / 1000.0f;
-    if (cur_time_cost > max_time_cost) {
-      max_time_cost = cur_time_cost;
-    }
-    if (cur_time_cost < min_time_cost) {
-      min_time_cost = cur_time_cost;
-    }
-    total_time_cost += cur_time_cost;
-    prediction_time = total_time_cost / REPEAT_COUNT;
-    printf("iter %d cost: %f ms\n", i, cur_time_cost);
-  }
-  printf("warmup: %d repeat: %d, average: %f ms, max: %f ms, min: %f ms\n",
-         WARMUP_COUNT, REPEAT_COUNT, prediction_time, max_time_cost,
-         min_time_cost);
-
-  // Get the data of output tensor and postprocess to output detected objects
-  std::unique_ptr<const paddle::lite_api::Tensor> output_tensor(
-      std::move(predictor->GetOutput(0)));
-  const float *output_data = output_tensor->mutable_data<float>();
-  int64_t output_size = 1;
-  for (auto dim : output_tensor->shape()) {
-    output_size *= dim;
-  }
-  cv::Mat output_image = input_image.clone();
-  double postprocess_start_time = get_current_us();
-  std::vector<RESULT> results =
-      postprocess(output_data, output_size, word_labels, SCORE_THRESHOLD,
-                  output_image, prediction_time);
-  double postprocess_end_time = get_current_us();
-  double postprocess_time =
-      (postprocess_end_time - postprocess_start_time) / 1000.0f;
-
-  printf("results: %d\n", results.size());
-  for (int i = 0; i < results.size(); i++) {
-    printf("[%d] %s - %f %f,%f,%f,%f\n", i, results[i].class_name.c_str(),
-           results[i].score, results[i].left, results[i].top, results[i].right,
-           results[i].bottom);
-  }
-  printf("Preprocess time: %f ms\n", preprocess_time);
-  printf("Prediction time: %f ms\n", prediction_time);
-  printf("Postprocess time: %f ms\n\n", postprocess_time);
-
-  return output_image;
-}
-
-int main(int argc, char **argv) {
-  if (argc < 5 || argc == 6) {
-    printf("Usage: \n"
-           "./object_detection_demo model_dir label_path [input_image_path] "
-           "[output_image_path]"
-           "use images from camera if input_image_path and input_image_path "
-           "isn't provided.");
-    return -1;
-  }
-
-  std::string model_path = argv[1];
-  std::string label_path = argv[2];
-  std::vector<std::string> word_labels = load_labels(label_path);
-  std::string nnadapter_subgraph_partition_config_path = argv[3];
-
-  std::string yaml_path = argv[4];
-  if (yaml_path != "null") {
-    load_yaml_config(yaml_path);
-  }
-
-  // Run inference by using full api with CxxConfig
-  paddle::lite_api::CxxConfig cxx_config;
-  if (1) { // combined model
-    cxx_config.set_model_file(model_path + "/model");
-    cxx_config.set_param_file(model_path + "/params");
-  } else {
-    cxx_config.set_model_dir(model_path);
-  }
-  cxx_config.set_threads(CPU_THREAD_NUM);
-  cxx_config.set_power_mode(CPU_POWER_MODE);
-
-  std::shared_ptr<paddle::lite_api::PaddlePredictor> predictor = nullptr;
-  std::vector<paddle::lite_api::Place> valid_places;
-  valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kARM), PRECISION(kInt8)});
-  valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kARM), PRECISION(kFloat)});
-  valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kNNAdapter), PRECISION(kInt8)});
-  valid_places.push_back(
-      paddle::lite_api::Place{TARGET(kNNAdapter), PRECISION(kFloat)});
-  cxx_config.set_valid_places(valid_places);
-  std::string device = "verisilicon_timvx";
-  cxx_config.set_nnadapter_device_names({device});
-  // cxx_config.set_nnadapter_context_properties(nnadapter_context_properties);
-
-  // cxx_config.set_nnadapter_model_cache_dir(nnadapter_model_cache_dir);
-  // Set the subgraph custom partition configuration file
-
-  if (!nnadapter_subgraph_partition_config_path.empty()) {
-    std::vector<char> nnadapter_subgraph_partition_config_buffer;
-    if (read_file(nnadapter_subgraph_partition_config_path,
-                  &nnadapter_subgraph_partition_config_buffer, false)) {
-      if (!nnadapter_subgraph_partition_config_buffer.empty()) {
-        std::string nnadapter_subgraph_partition_config_string(
-            nnadapter_subgraph_partition_config_buffer.data(),
-            nnadapter_subgraph_partition_config_buffer.size());
-        cxx_config.set_nnadapter_subgraph_partition_config_buffer(
-            nnadapter_subgraph_partition_config_string);
-      }
-    } else {
-      printf("Failed to load the subgraph custom partition configuration file "
-             "%s\n",
-             nnadapter_subgraph_partition_config_path.c_str());
-    }
-  }
-
-  try {
-    predictor = paddle::lite_api::CreatePaddlePredictor(cxx_config);
-    predictor->SaveOptimizedModel(
-        model_path, paddle::lite_api::LiteModelType::kNaiveBuffer);
-  } catch (std::exception e) {
-    printf("An internal error occurred in PaddleLite(cxx config).\n");
-  }
-
-  paddle::lite_api::MobileConfig config;
-  config.set_model_from_file(model_path + ".nb");
-  config.set_threads(CPU_THREAD_NUM);
-  config.set_power_mode(CPU_POWER_MODE);
-  config.set_nnadapter_device_names({device});
-  predictor =
-      paddle::lite_api::CreatePaddlePredictor<paddle::lite_api::MobileConfig>(
-          config);
-  if (argc > 5) {
-    WARMUP_COUNT = 1;
-    REPEAT_COUNT = 5;
-    std::string input_image_path = argv[5];
-    std::string output_image_path = argv[6];
-    cv::Mat input_image = cv::imread(input_image_path);
-    cv::Mat output_image = process(input_image, word_labels, predictor);
-    cv::imwrite(output_image_path, output_image);
-    cv::imshow("Object Detection Demo", output_image);
-    cv::waitKey(0);
-  } else {
-    cv::VideoCapture cap(1);
-    cap.set(cv::CAP_PROP_FRAME_WIDTH, 640);
-    cap.set(cv::CAP_PROP_FRAME_HEIGHT, 480);
-    if (!cap.isOpened()) {
-      return -1;
-    }
-    while (1) {
-      cv::Mat input_image;
-      cap >> input_image;
-      cv::Mat output_image = process(input_image, word_labels, predictor);
-      cv::imshow("Object Detection Demo", output_image);
-      if (cv::waitKey(1) == char('q')) {
-        break;
-      }
-    }
-    cap.release();
-    cv::destroyAllWindows();
-  }
-  return 0;
-}
--- a/examples/vision/detection/paddledetection/rk1126/picodet_detection/run.sh
+++ b/examples/vision/detection/paddledetection/rk1126/picodet_detection/run.sh
@@ -1,15 +0,0 @@
-#!/bin/bash
-#run
-
-TARGET_ABI=armv8 # for 64bit
-#TARGET_ABI=armv7hf # for 32bit
-if [ -n "$1" ]; then
-    TARGET_ABI=$1
-fi
-export LD_LIBRARY_PATH=../Paddle-Lite/libs/$TARGET_ABI/
-export GLOG_v=0
-export VSI_NN_LOG_LEVEL=0
-export VIV_VX_ENABLE_GRAPH_TRANSFORM=-pcq:1
-export VIV_VX_SET_PER_CHANNEL_ENTROPY=100
-export TIMVX_BATCHNORM_FUSION_MAX_ALLOWED_QUANT_SCALE_DEVIATION=30000
-build/object_detection_demo models/picodetv2_relu6_coco_no_fuse ../../assets/labels/coco_label_list.txt models/picodetv2_relu6_coco_no_fuse/subgraph.txt models/picodetv2_relu6_coco_no_fuse/picodet.yml 
--- a/examples/vision/detection/paddledetection/rv1126/README.md
+++ b/examples/vision/detection/paddledetection/rv1126/README.md
@@ -0,0 +1,11 @@
+# PP-YOLOE  量化模型在 RV1126 上的部署
+目前 FastDeploy 已经支持基于 PaddleLite 部署 PP-YOLOE  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)
+
+
+## 详细部署文档
+
+在 RV1126 上只支持 C++ 的部署。
+
+- [C++部署](cpp)
--- a/examples/vision/detection/paddledetection/rv1126/cpp/CMakeLists.txt
+++ b/examples/vision/detection/paddledetection/rv1126/cpp/CMakeLists.txt
@@ -0,0 +1,38 @@
+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})
+include_directories(${FastDeploy_INCLUDE_DIRS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer_ppyoloe.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
+
+set(CMAKE_INSTALL_PREFIX ${CMAKE_SOURCE_DIR}/build/install)
+
+install(TARGETS infer_demo DESTINATION ./)
+
+install(DIRECTORY models DESTINATION ./)
+install(DIRECTORY images DESTINATION ./)
+# install(DIRECTORY run_with_adb.sh DESTINATION ./)
+
+file(GLOB FASTDEPLOY_LIBS ${FASTDEPLOY_INSTALL_DIR}/lib/*)
+install(PROGRAMS ${FASTDEPLOY_LIBS} DESTINATION lib)
+
+file(GLOB OPENCV_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/opencv/lib/lib*)
+install(PROGRAMS ${OPENCV_LIBS} DESTINATION lib)
+
+file(GLOB PADDLELITE_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/paddlelite/lib/lib*)
+install(PROGRAMS ${PADDLELITE_LIBS} DESTINATION lib)
+
+file(GLOB TIMVX_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/paddlelite/lib/verisilicon_timvx/*)
+install(PROGRAMS ${TIMVX_LIBS} DESTINATION lib)
+
+file(GLOB ADB_TOOLS run_with_adb.sh)
+install(PROGRAMS ${ADB_TOOLS} DESTINATION ./)
--- a/examples/vision/detection/paddledetection/rv1126/cpp/README.md
+++ b/examples/vision/detection/paddledetection/rv1126/cpp/README.md
@@ -0,0 +1,55 @@
+# PP-YOLOE  量化模型 C++ 部署示例
+
+本目录下提供的 `infer.cc`，可以帮助用户快速完成 PP-YOLOE 量化模型在 RV1126 上的部署推理加速。
+
+## 部署准备
+### FastDeploy 交叉编译环境准备
+- 1. 软硬件环境满足要求，以及交叉编译环境的准备，请参考：[FastDeploy 交叉编译环境准备](../../../../../../docs/cn/build_and_install/rv1126.md#交叉编译环境搭建)  
+
+### 模型准备
+- 1. 用户可以直接使用由 FastDeploy 提供的量化模型进行部署。
+- 2. 用户可以先使用 PaddleDetection 自行导出 Float32 模型，注意导出模型模型时设置参数：use_shared_conv=False，更多细节请参考：[PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/ppyoloe)
+- 3. 用户可以使用 FastDeploy 提供的[一键模型自动化压缩工具](../../../../../../tools/common_tools/auto_compression/)，自行进行模型量化, 并使用产出的量化模型进行部署。（注意: 推理量化后的检测模型仍然需要FP32模型文件夹下的 infer_cfg.yml 文件，自行量化的模型文件夹内不包含此 yaml 文件，用户从 FP32 模型文件夹下复制此yaml文件到量化后的模型文件夹内即可。）
+- 更多量化相关相关信息可查阅[模型量化](../../quantize/README.md)
+
+## 在 RV1126 上部署量化后的 PP-YOLOE  检测模型
+请按照以下步骤完成在 RV1126 上部署 PP-YOLOE  量化模型：
+1. 交叉编译编译 FastDeploy 库，具体请参考：[交叉编译 FastDeploy](../../../../../../docs/cn/build_and_install/rv1126.md#基于-paddlelite-的-fastdeploy-交叉编译库编译)
+
+2. 将编译后的库拷贝到当前目录，可使用如下命令：
+```bash
+cp -r FastDeploy/build/fastdeploy-tmivx/ FastDeploy/examples/vision/detection/yolov5/rv1126/cpp
+```
+
+3. 在当前路径下载部署所需的模型和示例图片：
+```bash
+mkdir models && mkdir images
+wget https://bj.bcebos.com/fastdeploy/models/ppyoloe_noshare_qat.tar.gz
+tar -xvf ppyoloe_noshare_qat.tar.gz
+cp -r ppyoloe_noshare_qat models
+wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
+cp -r 000000014439.jpg images
+```
+
+4. 编译部署示例，可使入如下命令：
+```bash
+mkdir build && cd build
+cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../fastdeploy-tmivx/timvx.cmake -DFASTDEPLOY_INSTALL_DIR=${PWD}/../fastdeploy-tmivx ..
+make -j8
+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：
+```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
+```
+
+部署成功后运行结果如下：
+
+<img width="640" src="https://user-images.githubusercontent.com/30516196/203708564-43c49485-9b48-4eb2-8fe7-0fa517979fff.png">
+
+需要特别注意的是，在 RV1126 上部署的模型需要是量化后的模型，模型的量化请参考：[模型量化](../../../../../../docs/cn/quantize.md)
--- a/examples/vision/detection/paddledetection/rv1126/cpp/infer_ppyoloe.cc
+++ b/examples/vision/detection/paddledetection/rv1126/cpp/infer_ppyoloe.cc
@@ -0,0 +1,66 @@
+// 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::RuntimeOption option;
+  option.UseTimVX();
+  option.SetLiteSubgraphPartitionPath(subgraph_file);
+
+  auto model = fastdeploy::vision::detection::PPYOLOE(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, 0.5);
+  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 "
+                 "path/to/image "
+                 "run_option, "
+                 "e.g ./infer_demo ./PPYOLOE_L_quant ./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;
+}
--- a/examples/vision/detection/paddledetection/rv1126/cpp/run_with_adb.sh
+++ b/examples/vision/detection/paddledetection/rv1126/cpp/run_with_adb.sh
@@ -0,0 +1,47 @@
+#!/bin/bash
+HOST_SPACE=${PWD}
+echo ${HOST_SPACE}
+WORK_SPACE=/data/local/tmp/test
+
+# The first parameter represents the demo name
+DEMO_NAME=image_classification_demo
+if [ -n "$1" ]; then
+  DEMO_NAME=$1
+fi
+
+# The second parameter represents the model name
+MODEL_NAME=mobilenet_v1_fp32_224
+if [ -n "$2" ]; then
+  MODEL_NAME=$2
+fi
+
+# The third parameter indicates the name of the image to be tested
+IMAGE_NAME=0001.jpg
+if [ -n "$3" ]; then
+  IMAGE_NAME=$3
+fi
+
+# The fourth parameter represents the ID of the device
+ADB_DEVICE_NAME=
+if [ -n "$4" ]; then
+  ADB_DEVICE_NAME="-s $4"
+fi
+
+# Set the environment variables required during the running process
+EXPORT_ENVIRONMENT_VARIABLES="export GLOG_v=5; export SUBGRAPH_ONLINE_MODE=true; export RKNPU_LOGLEVEL=5; export RKNN_LOG_LEVEL=5; ulimit -c unlimited; export VIV_VX_ENABLE_GRAPH_TRANSFORM=-pcq:1; export VIV_VX_SET_PER_CHANNEL_ENTROPY=100; export TIMVX_BATCHNORM_FUSION_MAX_ALLOWED_QUANT_SCALE_DEVIATION=300000; export VSI_NN_LOG_LEVEL=5;"
+
+EXPORT_ENVIRONMENT_VARIABLES="${EXPORT_ENVIRONMENT_VARIABLES}export LD_LIBRARY_PATH=${WORK_SPACE}/lib:\$LD_LIBRARY_PATH;"
+
+# Please install adb, and DON'T run this in the docker.
+set -e
+adb $ADB_DEVICE_NAME shell "rm -rf $WORK_SPACE"
+adb $ADB_DEVICE_NAME shell "mkdir -p $WORK_SPACE"
+
+# Upload the demo, librarys, model and test images to the device
+adb $ADB_DEVICE_NAME push ${HOST_SPACE}/lib $WORK_SPACE
+adb $ADB_DEVICE_NAME push ${HOST_SPACE}/${DEMO_NAME} $WORK_SPACE
+adb $ADB_DEVICE_NAME push models $WORK_SPACE
+adb $ADB_DEVICE_NAME push images $WORK_SPACE
+
+# Execute the deployment demo
+adb $ADB_DEVICE_NAME shell "cd $WORK_SPACE; ${EXPORT_ENVIRONMENT_VARIABLES} chmod +x ./${DEMO_NAME}; ./${DEMO_NAME} ./models/${MODEL_NAME} ./images/$IMAGE_NAME"
--- a/examples/vision/detection/yolov5/quantize/README.md
+++ b/examples/vision/detection/yolov5/quantize/README.md
@@ -4,7 +4,7 @@ FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工

 ## FastDeploy一键模型自动化压缩工具
 FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
-详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/common_tools/auto_compression/)

 ## 下载量化完成的YOLOv5s模型
 用户也可以直接下载下表中的量化模型进行部署.(点击模型名字即可下载)
--- a/examples/vision/detection/yolov5/quantize/README_EN.md
+++ b/examples/vision/detection/yolov5/quantize/README_EN.md
@@ -6,7 +6,7 @@ Users can use the one-click model quantization tool to quantize and deploy the m
 ## FastDeploy One-Click Model Quantization Tool

 FastDeploy provides a one-click quantization tool that allows users to quantize a model simply with a configuration file.
-For a detailed tutorial, please refer to: [One-Click Model Quantization Tool](../../../../../tools/auto_compression/)
+For a detailed tutorial, please refer to: [One-Click Model Quantization Tool](../../../../../tools/common_tools/auto_compression/)

 ## Download Quantized YOLOv5s Model

--- a/examples/vision/detection/yolov5/quantize/cpp/README.md
+++ b/examples/vision/detection/yolov5/quantize/cpp/README.md
@@ -9,7 +9,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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.
+- 2. 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.

 ## 以量化后的YOLOv5s模型为例, 进行部署
 在本目录执行如下命令即可完成编译,以及量化模型部署.支持此模型需保证FastDeploy版本0.7.0以上(x.x.x>=0.7.0)
--- a/examples/vision/detection/yolov5/quantize/python/README.md
+++ b/examples/vision/detection/yolov5/quantize/python/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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.
+- 2. 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.


 ## 以量化后的YOLOv5s模型为例, 进行部署
--- a/examples/vision/detection/yolov5/rv1126/README.md
+++ b/examples/vision/detection/yolov5/rv1126/README.md
@@ -0,0 +1,11 @@
+# YOLOv5 量化模型在 RV1126 上的部署
+目前 FastDeploy 已经支持基于 PaddleLite 部署 YOLOv5 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)
+
+
+## 详细部署文档
+
+在 RV1126 上只支持 C++ 的部署。
+
+- [C++部署](cpp)
--- a/examples/vision/detection/yolov5/rv1126/cpp/CMakeLists.txt
+++ b/examples/vision/detection/yolov5/rv1126/cpp/CMakeLists.txt
@@ -0,0 +1,37 @@
+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})
+include_directories(${FastDeploy_INCLUDE_DIRS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
+
+set(CMAKE_INSTALL_PREFIX ${CMAKE_SOURCE_DIR}/build/install)
+
+install(TARGETS infer_demo DESTINATION ./)
+
+install(DIRECTORY models DESTINATION ./)
+install(DIRECTORY images DESTINATION ./)
+
+file(GLOB FASTDEPLOY_LIBS ${FASTDEPLOY_INSTALL_DIR}/lib/*)
+install(PROGRAMS ${FASTDEPLOY_LIBS} DESTINATION lib)
+
+file(GLOB OPENCV_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/opencv/lib/lib*)
+install(PROGRAMS ${OPENCV_LIBS} DESTINATION lib)
+
+file(GLOB PADDLELITE_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/paddlelite/lib/lib*)
+install(PROGRAMS ${PADDLELITE_LIBS} DESTINATION lib)
+
+file(GLOB TIMVX_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/paddlelite/lib/verisilicon_timvx/*)
+install(PROGRAMS ${TIMVX_LIBS} DESTINATION lib)
+
+file(GLOB ADB_TOOLS run_with_adb.sh)
+install(PROGRAMS ${ADB_TOOLS} DESTINATION ./)
--- a/examples/vision/detection/yolov5/rv1126/cpp/README.md
+++ b/examples/vision/detection/yolov5/rv1126/cpp/README.md
@@ -0,0 +1,54 @@
+# YOLOv5 量化模型 C++ 部署示例
+
+本目录下提供的 `infer.cc`，可以帮助用户快速完成 YOLOv5 量化模型在 RV1126 上的部署推理加速。
+
+## 部署准备
+### FastDeploy 交叉编译环境准备
+- 1. 软硬件环境满足要求，以及交叉编译环境的准备，请参考：[FastDeploy 交叉编译环境准备](../../../../../../docs/cn/build_and_install/rv1126.md#交叉编译环境搭建)  
+
+### 量化模型准备
+- 1. 用户可以直接使用由 FastDeploy 提供的量化模型进行部署。
+- 2. 用户可以使用 FastDeploy 提供的[一键模型自动化压缩工具](../../../../../../tools/common_tools/auto_compression/),自行进行模型量化, 并使用产出的量化模型进行部署。
+- 更多量化相关相关信息可查阅[模型量化](../../quantize/README.md)
+
+## 在 RV1126 上部署量化后的 YOLOv5 检测模型
+请按照以下步骤完成在 RV1126 上部署 YOLOv5 量化模型：
+1. 交叉编译编译 FastDeploy 库，具体请参考：[交叉编译 FastDeploy](../../../../../../docs/cn/build_and_install/rv1126.md#基于-paddlelite-的-fastdeploy-交叉编译库编译)
+
+2. 将编译后的库拷贝到当前目录，可使用如下命令：
+```bash
+cp -r FastDeploy/build/fastdeploy-tmivx/ FastDeploy/examples/vision/detection/yolov5/rv1126/cpp
+```
+
+3. 在当前路径下载部署所需的模型和示例图片：
+```bash
+mkdir models && mkdir images
+wget https://bj.bcebos.com/fastdeploy/models/yolov5s_ptq_model.tar.gz
+tar -xvf yolov5s_ptq_model.tar.gz
+cp -r yolov5s_ptq_model models
+wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
+cp -r 000000014439.jpg images
+```
+
+4. 编译部署示例，可使入如下命令：
+```bash
+mkdir build && cd build
+cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../fastdeploy-tmivx/timvx.cmake -DFASTDEPLOY_INSTALL_DIR=${PWD}/../fastdeploy-tmivx ..
+make -j8
+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 YOLOv5 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/yolov5/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 yolov5s_ptq_model 000000014439.jpg $DEVICE_ID
+```
+
+部署成功后，vis_result.jpg 保存的结果如下：
+
+<img width="640" src="https://user-images.githubusercontent.com/30516196/203706969-dd58493c-6635-4ee7-9421-41c2e0c9524b.png">
+
+需要特别注意的是，在 RV1126 上部署的模型需要是量化后的模型，模型的量化请参考：[模型量化](../../../../../../docs/cn/quantize.md)
--- a/examples/vision/detection/yolov5/rv1126/cpp/infer.cc
+++ b/examples/vision/detection/yolov5/rv1126/cpp/infer.cc
@@ -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 subgraph_file = model_dir + sep + "subgraph.txt";
+
+  fastdeploy::RuntimeOption option;
+  option.UseTimVX();
+  option.SetLiteSubgraphPartitionPath(subgraph_file);
+
+  auto model = fastdeploy::vision::detection::YOLOv5(
+      model_file, params_file, option, fastdeploy::ModelFormat::PADDLE);
+  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);
+  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 "
+                 "path/to/image "
+                 "run_option, "
+                 "e.g ./infer_demo ./yolov5s_quant ./000000014439.jpg"
+              << std::endl;
+    return -1;
+  }
+
+  std::string model_dir = argv[1];
+  std::string test_image = argv[2];
+  InitAndInfer(model_dir, test_image);
+  return 0;
+}
--- a/examples/vision/detection/yolov5/rv1126/cpp/run_with_adb.sh
+++ b/examples/vision/detection/yolov5/rv1126/cpp/run_with_adb.sh
@@ -0,0 +1,47 @@
+#!/bin/bash
+HOST_SPACE=${PWD}
+echo ${HOST_SPACE}
+WORK_SPACE=/data/local/tmp/test
+
+# The first parameter represents the demo name
+DEMO_NAME=image_classification_demo
+if [ -n "$1" ]; then
+  DEMO_NAME=$1
+fi
+
+# The second parameter represents the model name
+MODEL_NAME=mobilenet_v1_fp32_224
+if [ -n "$2" ]; then
+  MODEL_NAME=$2
+fi
+
+# The third parameter indicates the name of the image to be tested
+IMAGE_NAME=0001.jpg
+if [ -n "$3" ]; then
+  IMAGE_NAME=$3
+fi
+
+# The fourth parameter represents the ID of the device
+ADB_DEVICE_NAME=
+if [ -n "$4" ]; then
+  ADB_DEVICE_NAME="-s $4"
+fi
+
+# Set the environment variables required during the running process
+EXPORT_ENVIRONMENT_VARIABLES="export GLOG_v=5; export VIV_VX_ENABLE_GRAPH_TRANSFORM=-pcq:1; export VIV_VX_SET_PER_CHANNEL_ENTROPY=100; export TIMVX_BATCHNORM_FUSION_MAX_ALLOWED_QUANT_SCALE_DEVIATION=300000; export VSI_NN_LOG_LEVEL=5;"
+
+EXPORT_ENVIRONMENT_VARIABLES="${EXPORT_ENVIRONMENT_VARIABLES}export LD_LIBRARY_PATH=${WORK_SPACE}/lib:\$LD_LIBRARY_PATH;"
+
+# Please install adb, and DON'T run this in the docker.
+set -e
+adb $ADB_DEVICE_NAME shell "rm -rf $WORK_SPACE"
+adb $ADB_DEVICE_NAME shell "mkdir -p $WORK_SPACE"
+
+# Upload the demo, librarys, model and test images to the device
+adb $ADB_DEVICE_NAME push ${HOST_SPACE}/lib $WORK_SPACE
+adb $ADB_DEVICE_NAME push ${HOST_SPACE}/${DEMO_NAME} $WORK_SPACE
+adb $ADB_DEVICE_NAME push models $WORK_SPACE
+adb $ADB_DEVICE_NAME push images $WORK_SPACE
+
+# Execute the deployment demo
+adb $ADB_DEVICE_NAME shell "cd $WORK_SPACE; ${EXPORT_ENVIRONMENT_VARIABLES} chmod +x ./${DEMO_NAME}; ./${DEMO_NAME} ./models/${MODEL_NAME} ./images/$IMAGE_NAME"
--- a/examples/vision/detection/yolov6/quantize/README.md
+++ b/examples/vision/detection/yolov6/quantize/README.md
@@ -4,7 +4,7 @@ FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工

 ## FastDeploy一键模型自动化压缩工具
 FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
-详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/common_tools/auto_compression/)
 ## 下载量化完成的YOLOv6s模型
 用户也可以直接下载下表中的量化模型进行部署.(点击模型名字即可下载)

--- a/examples/vision/detection/yolov6/quantize/README_EN.md
+++ b/examples/vision/detection/yolov6/quantize/README_EN.md
@@ -6,7 +6,7 @@ Users can use the one-click model quantization tool to quantize and deploy the m
 ## FastDeploy One-Click Model Quantization Tool

 FastDeploy provides a one-click quantization tool that allows users to quantize a model simply with a configuration file.
-For detailed tutorial, please refer to : [One-Click Model Quantization Tool](../../../../../tools/auto_compression/)
+For detailed tutorial, please refer to : [One-Click Model Quantization Tool](../../../../../tools/common_tools/auto_compression/)

 ## Download Quantized YOLOv6s Model

--- a/examples/vision/detection/yolov6/quantize/cpp/README.md
+++ b/examples/vision/detection/yolov6/quantize/cpp/README.md
@@ -9,7 +9,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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.
+- 2. 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.

 ## 以量化后的YOLOv6s模型为例, 进行部署
 在本目录执行如下命令即可完成编译,以及量化模型部署.支持此模型需保证FastDeploy版本0.7.0以上(x.x.x>=0.7.0)
--- a/examples/vision/detection/yolov6/quantize/python/README.md
+++ b/examples/vision/detection/yolov6/quantize/python/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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.
+- 2. 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.

 ## 以量化后的YOLOv6s模型为例, 进行部署
 ```bash
--- a/examples/vision/detection/yolov7/quantize/README.md
+++ b/examples/vision/detection/yolov7/quantize/README.md
@@ -4,7 +4,7 @@ FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工

 ## FastDeploy一键模型自动化压缩工具
 FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
-详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/common_tools/auto_compression/)

 ## 下载量化完成的YOLOv7模型
 用户也可以直接下载下表中的量化模型进行部署.(点击模型名字即可下载)
--- a/examples/vision/detection/yolov7/quantize/README_EN.md
+++ b/examples/vision/detection/yolov7/quantize/README_EN.md
@@ -6,7 +6,7 @@ Users can use the one-click model quantization tool to quantize and deploy the m
 ## FastDeploy One-Click Model Quantization Tool

 FastDeploy provides a one-click quantization tool that allows users to quantize a model simply with a configuration file.
-For detailed tutorial, please refer to : [One-Click Model Quantization Tool](../../../../../tools/auto_compression/)
+For detailed tutorial, please refer to : [One-Click Model Quantization Tool](../../../../../tools/common_tools/auto_compression/)

 ## Download Quantized YOLOv7 Model

--- a/examples/vision/detection/yolov7/quantize/cpp/README.md
+++ b/examples/vision/detection/yolov7/quantize/cpp/README.md
@@ -9,7 +9,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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.
+- 2. 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.

 ## 以量化后的YOLOv7模型为例, 进行部署
 在本目录执行如下命令即可完成编译,以及量化模型部署.支持此模型需保证FastDeploy版本0.7.0以上(x.x.x>=0.7.0)
--- a/examples/vision/detection/yolov7/quantize/python/README.md
+++ b/examples/vision/detection/yolov7/quantize/python/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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.
+- 2. 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.

 ## 以量化后的YOLOv7模型为例, 进行部署
 ```bash
--- a/examples/vision/segmentation/paddleseg/quantize/README.md
+++ b/examples/vision/segmentation/paddleseg/quantize/README.md
@@ -4,7 +4,7 @@ FastDeploy已支持部署量化模型,并提供一键模型自动化压缩的工

 ## FastDeploy一键模型自动化压缩工具
 FastDeploy 提供了一键模型自动化压缩工具, 能够简单地通过输入一个配置文件, 对模型进行量化.
-详细教程请见: [一键模型自动化压缩工具](../../../../../tools/auto_compression/)
+详细教程请见: [一键模型自动化压缩工具](../../../../../tools/common_tools/auto_compression/)
 注意: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml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。

 ## 下载量化完成的PaddleSeg模型
--- a/examples/vision/segmentation/paddleseg/quantize/cpp/README.md
+++ b/examples/vision/segmentation/paddleseg/quantize/cpp/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml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.)
+- 2. 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: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml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.)

 ## 以量化后的PP_LiteSeg_T_STDC1_cityscapes模型为例, 进行部署
 在本目录执行如下命令即可完成编译,以及量化模型部署.支持此模型需保证FastDeploy版本0.7.0以上(x.x.x>=0.7.0)
--- a/examples/vision/segmentation/paddleseg/quantize/python/README.md
+++ b/examples/vision/segmentation/paddleseg/quantize/python/README.md
@@ -8,7 +8,7 @@

 ### 量化模型准备
 - 1. 用户可以直接使用由FastDeploy提供的量化模型进行部署.
- 2. 用户可以使用FastDeploy提供的[一键模型自动化压缩工具](../../../../../../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: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml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.)
+- 2. 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: 推理量化后的分类模型仍然需要FP32模型文件夹下的deploy.yaml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.)


 ## 以量化后的PP_LiteSeg_T_STDC1_cityscapes模型为例, 进行部署
--- a/examples/vision/segmentation/paddleseg/rv1126/README.md
+++ b/examples/vision/segmentation/paddleseg/rv1126/README.md
@@ -0,0 +1,11 @@
+# PP-LiteSeg 量化模型在 RV1126 上的部署
+目前 FastDeploy 已经支持基于 PaddleLite 部署 PP-LiteSeg 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)
+
+
+## 详细部署文档
+
+在 RV1126 上只支持 C++ 的部署。
+
+- [C++部署](cpp)
--- a/examples/vision/segmentation/paddleseg/rv1126/cpp/CMakeLists.txt
+++ b/examples/vision/segmentation/paddleseg/rv1126/cpp/CMakeLists.txt
@@ -0,0 +1,38 @@
+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})
+include_directories(${FastDeploy_INCLUDE_DIRS})
+
+add_executable(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
+
+set(CMAKE_INSTALL_PREFIX ${CMAKE_SOURCE_DIR}/build/install)
+
+install(TARGETS infer_demo DESTINATION ./)
+
+install(DIRECTORY models DESTINATION ./)
+install(DIRECTORY images DESTINATION ./)
+# install(DIRECTORY run_with_adb.sh DESTINATION ./)
+
+file(GLOB FASTDEPLOY_LIBS ${FASTDEPLOY_INSTALL_DIR}/lib/*)
+install(PROGRAMS ${FASTDEPLOY_LIBS} DESTINATION lib)
+
+file(GLOB OPENCV_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/opencv/lib/lib*)
+install(PROGRAMS ${OPENCV_LIBS} DESTINATION lib)
+
+file(GLOB PADDLELITE_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/paddlelite/lib/lib*)
+install(PROGRAMS ${PADDLELITE_LIBS} DESTINATION lib)
+
+file(GLOB TIMVX_LIBS ${FASTDEPLOY_INSTALL_DIR}/third_libs/install/paddlelite/lib/verisilicon_timvx/*)
+install(PROGRAMS ${TIMVX_LIBS} DESTINATION lib)
+
+file(GLOB ADB_TOOLS run_with_adb.sh)
+install(PROGRAMS ${ADB_TOOLS} DESTINATION ./)
--- a/examples/vision/segmentation/paddleseg/rv1126/cpp/README.md
+++ b/examples/vision/segmentation/paddleseg/rv1126/cpp/README.md
@@ -0,0 +1,54 @@
+# PP-LiteSeg 量化模型 C++ 部署示例
+
+本目录下提供的 `infer.cc`，可以帮助用户快速完成 PP-LiteSeg 量化模型在 RV1126 上的部署推理加速。
+
+## 部署准备
+### FastDeploy 交叉编译环境准备
+- 1. 软硬件环境满足要求，以及交叉编译环境的准备，请参考：[FastDeploy 交叉编译环境准备](../../../../../../docs/cn/build_and_install/rv1126.md#交叉编译环境搭建)  
+
+### 模型准备
+- 1. 用户可以直接使用由 FastDeploy 提供的量化模型进行部署。
+- 2. 用户可以使用 FastDeploy 提供的一键模型自动化压缩工具,自行进行模型量化, 并使用产出的量化模型进行部署.(注意: 推理量化后的分类模型仍然需要FP32模型文件夹下的 deploy.yaml 文件, 自行量化的模型文件夹内不包含此 yaml 文件, 用户从FP32模型文件夹下复制此yaml文件到量化后的模型文件夹内即可.)
+- 更多量化相关相关信息可查阅[模型量化](../../quantize/README.md)
+
+## 在 RV1126 上部署量化后的 PP-LiteSeg 分割模型
+请按照以下步骤完成在 RV1126 上部署 PP-LiteSeg 量化模型：
+1. 交叉编译编译 FastDeploy 库，具体请参考：[交叉编译 FastDeploy](../../../../../../docs/cn/build_and_install/rv1126.md#基于-paddlelite-的-fastdeploy-交叉编译库编译)
+
+2. 将编译后的库拷贝到当前目录，可使用如下命令：
+```bash
+cp -r FastDeploy/build/fastdeploy-tmivx/ FastDeploy/examples/vision/segmentation/paddleseg/rv1126/cpp
+```
+
+3. 在当前路径下载部署所需的模型和示例图片：
+```bash
+mkdir models && mkdir images
+wget https://bj.bcebos.com/fastdeploy/models/rk1/ppliteseg.tar.gz
+tar -xvf ppliteseg.tar.gz
+cp -r ppliteseg models
+wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png
+cp -r cityscapes_demo.png images
+```
+
+4. 编译部署示例，可使入如下命令：
+```bash
+mkdir build && cd build
+cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../fastdeploy-tmivx/timvx.cmake -DFASTDEPLOY_INSTALL_DIR=${PWD}/../fastdeploy-tmivx ..
+make -j8
+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-LiteSeg 分割模型到 Rockchip RV1126，可使用如下命令：
+```bash
+# 进入 install 目录
+cd FastDeploy/examples/vision/segmentation/paddleseg/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 ppliteseg cityscapes_demo.png $DEVICE_ID
+```
+
+部署成功后运行结果如下：
+
+<img width="640" src="https://user-images.githubusercontent.com/30516196/205544166-9b2719ff-ed82-4908-b90a-095de47392e1.png">
+
+需要特别注意的是，在 RV1126 上部署的模型需要是量化后的模型，模型的量化请参考：[模型量化](../../../../../../docs/cn/quantize.md)
--- a/examples/vision/segmentation/paddleseg/rv1126/cpp/infer.cc
+++ b/examples/vision/segmentation/paddleseg/rv1126/cpp/infer.cc
@@ -0,0 +1,66 @@
+// 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 + "deploy.yaml";
+  auto subgraph_file = model_dir + sep + "subgraph.txt";
+
+  fastdeploy::RuntimeOption option;
+  option.UseTimVX();
+  option.SetLiteSubgraphPartitionPath(subgraph_file);
+  
+  auto model = fastdeploy::vision::segmentation::PaddleSegModel(
+      model_file, params_file, config_file,option);
+
+  assert(model.Initialized());
+
+  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 < 3) {
+    std::cout << "Usage: infer_demo path/to/quant_model "
+                 "path/to/image "
+                 "run_option, "
+                 "e.g ./infer_demo ./ResNet50_vd_quant ./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;
+}
--- a/examples/vision/segmentation/paddleseg/rv1126/cpp/run_with_adb.sh
+++ b/examples/vision/segmentation/paddleseg/rv1126/cpp/run_with_adb.sh
@@ -0,0 +1,47 @@
+#!/bin/bash
+HOST_SPACE=${PWD}
+echo ${HOST_SPACE}
+WORK_SPACE=/data/local/tmp/test
+
+# The first parameter represents the demo name
+DEMO_NAME=image_classification_demo
+if [ -n "$1" ]; then
+  DEMO_NAME=$1
+fi
+
+# The second parameter represents the model name
+MODEL_NAME=mobilenet_v1_fp32_224
+if [ -n "$2" ]; then
+  MODEL_NAME=$2
+fi
+
+# The third parameter indicates the name of the image to be tested
+IMAGE_NAME=0001.jpg
+if [ -n "$3" ]; then
+  IMAGE_NAME=$3
+fi
+
+# The fourth parameter represents the ID of the device
+ADB_DEVICE_NAME=
+if [ -n "$4" ]; then
+  ADB_DEVICE_NAME="-s $4"
+fi
+
+# Set the environment variables required during the running process
+EXPORT_ENVIRONMENT_VARIABLES="export GLOG_v=5; export VIV_VX_ENABLE_GRAPH_TRANSFORM=-pcq:1; export VIV_VX_SET_PER_CHANNEL_ENTROPY=100; export TIMVX_BATCHNORM_FUSION_MAX_ALLOWED_QUANT_SCALE_DEVIATION=300000; export VSI_NN_LOG_LEVEL=5;"
+
+EXPORT_ENVIRONMENT_VARIABLES="${EXPORT_ENVIRONMENT_VARIABLES}export LD_LIBRARY_PATH=${WORK_SPACE}/lib:\$LD_LIBRARY_PATH;"
+
+# Please install adb, and DON'T run this in the docker.
+set -e
+adb $ADB_DEVICE_NAME shell "rm -rf $WORK_SPACE"
+adb $ADB_DEVICE_NAME shell "mkdir -p $WORK_SPACE"
+
+# Upload the demo, librarys, model and test images to the device
+adb $ADB_DEVICE_NAME push ${HOST_SPACE}/lib $WORK_SPACE
+adb $ADB_DEVICE_NAME push ${HOST_SPACE}/${DEMO_NAME} $WORK_SPACE
+adb $ADB_DEVICE_NAME push models $WORK_SPACE
+adb $ADB_DEVICE_NAME push images $WORK_SPACE
+
+# Execute the deployment demo
+adb $ADB_DEVICE_NAME shell "cd $WORK_SPACE; ${EXPORT_ENVIRONMENT_VARIABLES} chmod +x ./${DEMO_NAME}; ./${DEMO_NAME} ./models/${MODEL_NAME} ./images/$IMAGE_NAME"
--- a/fastdeploy/backends/lite/lite_backend.cc
+++ b/fastdeploy/backends/lite/lite_backend.cc
@@ -72,7 +72,7 @@ void LiteBackend::BuildOption(const LiteBackendOption& option) {
      }
    }
  }
-  if(option_.enable_timvx){
+  if(option_.enable_timvx) {
    config_.set_nnadapter_device_names({"verisilicon_timvx"});
    valid_places.push_back(
          paddle::lite_api::Place{TARGET(kNNAdapter), PRECISION(kInt8)});
--- a/fastdeploy/vision/detection/contrib/yolov5/yolov5.cc
+++ b/fastdeploy/vision/detection/contrib/yolov5/yolov5.cc
@@ -27,6 +27,7 @@ YOLOv5::YOLOv5(const std::string& model_file, const std::string& params_file,
  } else {
    valid_cpu_backends = {Backend::PDINFER, Backend::ORT, Backend::LITE};
    valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
+    valid_timvx_backends = {Backend::LITE};
  }
  runtime_option = custom_option;
  runtime_option.model_format = model_format;
--- a/fastdeploy/vision/detection/ppdet/model.h
+++ b/fastdeploy/vision/detection/ppdet/model.h
@@ -64,6 +64,7 @@ class FASTDEPLOY_DECL PPYOLOE : public PPDetBase {
    valid_cpu_backends = {Backend::OPENVINO, Backend::ORT,
                        Backend::PDINFER, Backend::LITE};
    valid_gpu_backends = {Backend::ORT, Backend::PDINFER, Backend::TRT};
+    valid_timvx_backends = {Backend::LITE};
    initialized = Initialize();
  }

--- a/fastdeploy/vision/segmentation/ppseg/model.cc
+++ b/fastdeploy/vision/segmentation/ppseg/model.cc
@@ -27,6 +27,7 @@ PaddleSegModel::PaddleSegModel(const std::string& model_file,
  valid_cpu_backends = {Backend::OPENVINO, Backend::PDINFER, Backend::ORT, Backend::LITE};
  valid_gpu_backends = {Backend::PDINFER, Backend::ORT, Backend::TRT};
  valid_rknpu_backends = {Backend::RKNPU2};
+  valid_timvx_backends = {Backend::LITE};
  runtime_option = custom_option;
  runtime_option.model_format = model_format;
  runtime_option.model_file = model_file;