Skip to main content

Environment Installation

This chapter mainly introduces the complete deployment method of the D-Robotics algorithm toolchain development environment.

Instructions for Using Deliverables​

Before deploying the algorithm toolchain environment, please download the Embedded Application Development Sample Delivery Package provided by D-Robotics to the Linux development machine environment.

Description of the Directory Structure of the Sample Package Source Code​

Unpack the SDK source code package of the algorithm toolchain:

  tar -xvf Ai_Toolchain_Package-release-vX.X.X-OE-vX.X.X.tar.xz

The directory structure after unpacking is as follows:

  • ai_benchmark: This directory provides evaluation examples for common classification, detection, and segmentation models, including performance evaluation and accuracy evaluation.

  • horizon_runtime_sample: This directory provides on-board examples for fixed-point models.

  • package: This directory contains some basic libraries and components for running the released artifacts.

    1. The board folder contains the executable program at the board side.

    2. The host folder contains the environment dependencies, tool dependencies, and libdnn libraries and headers related to model inference in the x86 development environment.

Development Machine Deployment​

For the environment deployment of the development machine, D-Robotics supports Docker deployment.

Development Machine Preparation​

In order to use the algorithm toolchain smoothly, D-Robotics recommends that your selected development machine should meet the following requirements:

Hardware/Operating SystemRequirements
CPUCPU I3 or higher or the equivalent E3/E5 processor
Memory16GB or higher
GPU(optional)CUDA11.6, driver version Linux:>= 510.39.01*
Compatible graphics cards include but are not limited to:
1) GeForce RTX 3090
2) GeForce RTX 2080 Ti
3) NVIDIA TITAN V
4) Tesla V100S-PCIE-32GB
SystemUbuntu 20.04

Using Docker Environment​

To help you quickly use the algorithm toolchain, D-Robotics provides Docker images with complete development environments, which greatly simplifies the deployment process.

Before reading this section, we hope that Docker's basic environment has been preinstalled on your development machine. The Docker basic environment information required by D-Robotics is as follows:

After completing the Docker environment installation, you need to add the non-root user to the Docker user group. Refer to the following command:

  sudo groupadd docker
  sudo gpasswd -a ${USER} docker
  sudo systemctl restart docker  # CentOS7/Ubuntu
  # re-login

The address of the Docker image needed in this section is as follows:

The naming format of the image file is:

  • GPU version docker: openexplorer/ai_toolchain_ubuntu_20_x3j5_gpu:{version}
Tip

When executing the command, replace {version} with the latest version of the Docker image you obtained, for example: D-Robotics docker hub GPU Docker The current latest version is openexplorer/ai_toolchain_ubuntu_20_x3j5_gpu:v1.0.0.

For the local Docker image package version, you can contact D-Robotics Technical Support Team to obtain it.

A development machine does not necessarily have a GPU card. Generally, a CPU development machine is used to load Docker images for model conversion!

Before using each image file for the first time, you need to pull the image.

  • The command to pull the image is:

      docker pull openexplorer/ai_toolchain_ubuntu_20_x3j5_gpu:v1.0.0

Then run the Docker container by executing the following command.

  • For CPU development machine Docker container, execute the following command:- Execute the following command in the Docker container on the GPU development machine:

      // Run the Docker image command

      export version=v1.0.0

      export ai_toolchain_package_path=/home/users/xxx/ai_toolchain_package

      export dataset_path=/home/users/xxx/data/

      docker run -it --runtime=nvidia -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
        -e NVIDIA_VISIBLE_DEVICES=all --rm --shm-size="15g" \
        -v "$ai_toolchain_package_path":/open_explorer \
        -v "$dataset_path":/data \
        openexplorer/ai_toolchain_ubuntu_20_x3j5_gpu:"${version}"

  • On the GPU development machine, execute the following command within a Docker container:

    # Command to run the docker image

    export version=v1.0.0
    export ai_toolchain_package_path=/home/users/xxx/ai_toolchain_package
    export dataset_path=/home/users/xxx/data/

    docker run -it --runtime=nvidia \
        -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
        -e NVIDIA_VISIBLE_DEVICES=all \
        --rm \
        --shm-size="15g" \
        -v "$ai_toolchain_package_path":/open_explorer \
        -v "$dataset_path":/data \
        openexplorer/ai_toolchain_ubuntu_20_x3j5_gpu:"${version}"
Note

When executing the above command:

KITTI3D: https://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d

CULane: https://xingangpan.github.io/projects/CULane.html

nuScenes: https://www.nuscenes.org/nuscenes

You have successfully entered the complete algorithm toolchain development environment through the Docker image. You can type hb_mapper --help command to verify if you can get help information:

  [root@d67382e74eea open_explorer]# hb_mapper --help
  Usage: hb_mapper [OPTIONS] COMMAND [ARGS]...

    hb_mapper is an offline model transform tool provided by horizon.

  Options:
    --version  Show the version and exit.
    --help     Show this message and exit.

  Commands:
    checker    check whether the model meet the requirements.
    infer      inference and dump output feature as float vector.
    makertbin  transform caffe model to quantization model, generate runtime...

If the hb_mapper tool has output logs properly, it means that the environment has been successfully installed and deployed. Please go to the Board Deployment section for board environment installation.

Board Deployment​

Board deployment requires you to update the board image to the latest version according to the flashing instructions. Please refer to the Install OS section for the upgrade method. After the upgrade is completed, copy the relevant supplementary files to the development board.

Preparation of Supplementary Files​

Some supplementary tools of the algorithm toolchain are not included in the system image. These tools have been placed in the Ai_Toolchain_Package-release-vX.X.X-OE-vX.X.X/package/ package. Enter the Ai_Toolchain_Package-release-vX.X.X-OE-vX.X.X/package/board directory and execute the installation script. The command is as follows:

  // For RDK X3 development board, execute the command
  bash install_xj3.sh ${board_ip}

  // For RDK Ultra development board, execute the command
  bash install_ultra.sh ${board_ip}
Note其中，${board_ip} is the IP address you have set for the development board. Please ensure that the IP is accessible from the development machine.

After successful installation, restart the development board and execute hrt_model_exec on the development board to verify the successful installation.

Instruction for Version Control Tool​

This chapter mainly introduces the usage instructions for the ddk_vcs version control tool, in order to help developers understand the current version status of the algorithm toolchain dependencies in the development machine environment.

Tip

The version control tool is mainly used for debugging when errors occur during model PTQ conversion using the Docker environment. If the model conversion function is normal, you can skip reading this chapter.

The version control tool has the following functionalities:

  • ddk_vcs list;
  • ddk_vcs install;
  • ddk_vcs uninstall;
  • ddk_vcs patch;
  • ddk_vcs show;

ddk_vcs list​

ddk_vcs list is used to list the installed software packages.

When executing this command without any parameters, the result will display the information of the installed modules. The usage example is as follows:


  [horizon@gpu-dev067 ai_toolchain]$ ddk_vcs list
  Host package version: v2.0.3
  The following packages versions
  Platform        Package         Version MD5
  --------------- --------------- ------- -------------
  aarch_64        appsdk          032419  093e13b44e
  aarch_64        dnn             1.8.1g  aff0f6f4de
  x86_64_gcc5.4.0 dnn_x86         1.8.1g  e8e6bf9ed5
  x86             horizon-nn      0.13.3  origin:0.13.3
  x86             horizon-nn-gpu  0.13.3  origin:N/A
  x86             horizon-tc-ui   1.6.4   origin:1.6.4
  x86             hbdk            3.28.3  origin:3.28.3
Note

The origin information in the last few lines will be updated to the current version in the environment after each installation using the install script in the toolchain SDK package. It will not change when using ddk_vcs for installation, only the value of Version will change.

When using the -p parameter, it will display the available module versions that can be installed. You can install them using ddk_vcs install. The usage example is as follows:

Host package version: 1.5.1
The following packages versions
Platform        Local Package                 Version MD5
--------------- ----------- ------- ----------
aarch_64        appsdk_1.9.0.tar.gz           1.9.0   bf01140c9d
aarch_64        bpu_predict_1.10.2.tar.gz     1.10.2  5b6e5dd6c5
aarch_64        dnn_1.1.2a.tar.gz             1.1.2a  fdb5729f4f
x86_64_gcc5.4.0 bpu_predict_1.10.2.tar.gz     1.10.2  4dbdd980a7
x86_64_gcc5.4.0 dnn_x86_1.1.2a.tar.gz         1.1.2a  5bf5fcd4fe

ddk_vcs install​

ddk_vcs install is used to install packages. Users can install the corresponding module tar package directly using "ddk_vcs install". The specified platform needs to be specified during installation. The usage is as follows:


  [horizon@gpu-dev004]$ ddk_vcs install bpu_predict_1.10.2.tar.gz -p aarch_64
  bpu_predict installed successfully, version: 1.10.2, platform: aarch_64
  [horizon@gpu-dev067 ai_toolchain]$ ddk_vcs install hbdk-3.28.3-py3-none-linux_x86_64.whl  horizon_nn-0.13.3-py3-none-any.whl
  hbdk-3.28.3-py3-none-linux_x86_64.whl installed successfully
  horizon_nn-0.13.3-py3-none-any.whl installed successfully

After using "ddk_vcs list -p", users can obtain the version information of each module package in their current host package. Then, "ddk_vcs install" can be used to conveniently switch between different versions. The usage is as follows:


  [horizon@gpu-dev004]$ ddk_vcs install bpu_predict==1.7.2  --platform aarch_64
  bpu_predict installed successfully, version: 1.7.2, platform: aarch_64

If the corresponding version is not available locally, you can specify the installation package location for installation.

ddk_vcs uninstall​

ddk_vcs uninstall is used to uninstall a specific module. The usage is as follows:


  [horizon@gpu-dev004]$ ddk_vcs uninstall bpu_predict --platform aarch_64
  Start to uninstall modules, platform: aarch_64
  bpu_predict uninstalled successfully, version: 1.10.2, platform: aarch_64

ddk_vcs patch​

"ddk_vcs patch ddk_patch.tar.gz" is used to install pre-made patch packages. The usage is as follows:


  [horizon@gpu-dev004]$ ddk_vcs patch ddk_patch.tar.gz
  bpu_predict installed successfully, version: 1.7.2_patch0, platform: aarch64

ddk_vcs show​

ddk_vcs show is used to display information about the installed packages. Use ddk_vcs show [module name] to display the information of the corresponding module. The usage is as follows:


  [horizon@gpu-dev004]$ ddk_vcs show bpu_predict
  Host package version 1.5.1
  The following packages versions
  Platform        Package     Version       MD5
  --------------- ----------- ------------- ----------
  aarch_64        bpu_predict 1.10.2        5b6e5dd6c5
  x86_64_gcc5.4.0 bpu_predict 1.10.2_patch1 d4f8e37921

If there are dependencies with the same name in two architectures, the -p/--platform option can be used to filter by architecture name. The usage is as follows:


  [horizon@gpu-dev004]$ ddk_vcs show bpu_predict -p aarch_64
  Host package version 1.5.1
  The following packages versions
  Platform Package     Version MD5
  -------- ----------- ------- ----------
  aarch_64 bpu_predict 1.10.2  5b6e5dd6c5