# ait_transplt_demos

**Repository Path**: wangguowei33/ait_transplt_demos

## Basic Information

- **Project Name**: ait_transplt_demos
- **Description**: 分别使用 OpenCV / MXBase 的 ResNet50 / YOLOV5 推理示例
- **Primary Language**: Python
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2023-07-11
- **Last Updated**: 2023-08-01

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 简介
- 分别使用 OpenCV / MXBase 的 ResNet50 / YOLOV5 推理示例
***

## 导出 onnx
  ```py
  import torchvision, torch
  mm = torchvision.models.resnet50(pretrained=True)
  torch.onnx.export(mm, torch.ones([1, 3, 224, 224]), 'resnet50.onnx')
  ```
  如需要可下载
  - [resnet50.onnx](https://gitee.com/wangguowei33/resnet_opencv_cpp/releases/download/onnx/resnet50.onnx)
  - [yolov5s.onnx](https://gitee.com/wangguowei33/resnet_opencv_cpp/releases/download/onnx/yolov5s.onnx)
## OpenCV ResNet50 推理
- 相关依赖
  ```sh
  sudo apt install libopencv-dev
  ```
  如安装失败，可使用源码编译
  ```sh
  git clone https://github.com/opencv/opencv.git
  cd opencv
  mkdir build && cd build && cmake .. && make
  cd -
  ```
- 编译执行
  ```sh
  cd resnet_opencv
  g++ -O3 resnet_opencv.cpp -o resnet_opencv `pkg-config --cflags --libs opencv4`
  ./resnet_opencv
  # Image process time duration: 4.31935ms
  # Model inference time duration: 230.819ms
  # index: 285
  # class: Egyptian_cat
  # score: 18.4915

  cd ../
  ```
## MXBase ResNet50 推理 + OpenCV 图像处理
- [MindX SDK 社区版](https://www.hiascend.com/zh/software/mindx-sdk/community) 下载安装 `Ascend-mindxsdk-mxvision`.
  ```sh
  ./Ascend-mindxsdk-mxvision_5.0.RC2_linux-aarch64.run --install
  source mxVision/set_env.sh
  ```
- 输出 om 模型
  ```sh
  cd assets
  SOC_VERSION=`python3 -c 'import acl; print(acl.get_soc_name())'`  # Ascend310 / Ascend310P3 or others
  atc --model resnet50.onnx --output resnet50 --framework 5 --soc_version $SOC_VERSION
  cd -
  ```
- 编译执行
  ```sh
  cd resnet_mxbase
  mkdir -p build && cd build && cmake .. && cd -
  cd build/ && make && cd - && ./resnet_mxbase
  # Image process time duration: 8.45735ms
  # Model inference time duration: 3.30807ms
  # index: 285
  # class: Egyptian_cat
  # score: 18.4844

  cd ../
  ```
## MXBase ResNet50 推理 + AIPP 数据处理
- [MindX SDK 社区版](https://www.hiascend.com/zh/software/mindx-sdk/community) 下载安装 `Ascend-mindxsdk-mxvision`.
  ```sh
  ./Ascend-mindxsdk-mxvision_5.0.RC2_linux-aarch64.run --install
  source mxVision/set_env.sh
  ```
- 输出 om 模型
  ```sh
  cd assets
  SOC_VERSION=`python3 -c 'import acl; print(acl.get_soc_name())'`  # Ascend310 / Ascend310P3 or others
  atc --model resnet50.onnx --output resnet50_aipp --framework 5 --soc_version $SOC_VERSION --insert_op_conf aipp_resnet.config
  cd -
  ```
- 编译执行
  ```sh
  cd resnet_mxbase_aipp
  mkdir -p build && cd build && cmake .. && cd -
  cd build/ && make && cd - && ./resnet_mxbase_aipp
  # Image process time duration: 16.2752ms
  # Model inference time duration: 2.8023ms
  # index: 285
  # class: Egyptian_cat
  # score: 18.5625

  cd ../
  ```