ApraPipes

A pipeline framework for developing video and image processing applications. Supports multiple GPUs and Machine Learning toolkits. More details can be found here https://apra-labs.github.io/ApraPipes.

Build status

Automatically built and tested on Ubuntu 18.04, Jetson Boards and Windows 11 x64 Visual Studio 2017 Community (without CUDA)

OS	Version	With Cuda	Tests	Status
Windows	2019	No	Yes
Ubuntu x64_86	20.04	No	Yes
Ubuntu ARM64 (Jetsons)	18.04	Yes	Yes
Ubuntu x64_86	18.04	Yes	No

Setup

Prerequisites for CUDA

• Make account on developer.nvidia.com, else the next steps will show HTTP 404/403 errors
• Download and install CUDA tool kit based on your OS:
  • Ubuntu 18.04 : CUDA Toolkit 10.2
  • Windows 10/11 : Cuda Toolkit 10.2
• Download Cudnn and extract where cuda is installed. Note: this is a painful process. Here are the steps:
  • Download the correct tar/zip file matching your cuda version. Do not download the exe/installer/deb package.
  • Windows:
    • download this file.
    • Extract the downloaded file and copy files to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2 using an administrative command prompt as follows

      cd .\extracted_folder
      cd include
      copy *.h "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\include\"
      cd ..\lib
      copy *.lib "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\lib\x64\"
      cd ..\bin
      copy *.dll "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\bin\"
      

  • Linux:
    • download this file
    • extract the files

      xz -d cudnn-linux-x86_64-8.3.2.44_cuda10.2-archive.tar.xz
      tar xvf cudnn-linux-x86_64-8.3.2.44_cuda10.2-archive.tar
      

    • copy files retaining the links

      cd ./cudnn-linux-x86_64-8.3.2.44_cuda10.2-archive
      sudo cp -P include/* /usr/local/cuda/include/
      sudo cp -P lib/* /usr/local/cuda/lib64/
      

Windows

Prerequisites

• Install Visual Studio 2017 Community

  • Install Desktop development C++
  • .NET Desktop development
  • Universal Windows Development Platform

• Install CMake 3.22.1

• Make sure git and git-lfs is installed

• Clone with submodules and LFS.

  git clone --recursive https://github.com/Apra-Labs/ApraPipes.git
  

• Run bootstrap-vcpkg.bat in the vcpkg/ directory

• Run

  vcpkg.exe integrate install
  

• Build Gstreamer

  cd thirdparty
  build_gstreamer.bat
  cd ..\..\..
  

  Note

  While building gstreamer:

  • If you get the error: python not found. Download Python 3.7.3 from here
  • If you get the error: meson not found. Install meson using python -m pip install meson

Build for windows

Without Cuda

If your windows machies does not have a GPU use this script

build_windows_no_cuda.bat

Note

If you get error: Could not find pkgConfig. Do the following:

1. Install choco. Open Windows PowerShell as Administrator and run:

Set-ExecutionPolicy AllSigned
Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; iex ((New-Object System.Net.WebClient).DownloadString('https://chocolatey.org/install.ps1'))

2. Run choco install pkgconfiglite

With Cuda

build_windows_cuda.bat

Run Tests

• list all tests

  _build/BUILD_TYPE/aprapipesut.exe --list_content
  

• run all tests

  _build/BUILD_TYPE/aprapipesut.exe
  

• run one test

  _build/BUILD_TYPE/aprapipesut.exe --run_test=filenamestrategy_tests/boostdirectorystrategy
  

• run one test with arguments

  _build/BUILD_TYPE/aprapipesut.exe --run_test=unit_tests/params_test -- -ip 10.102.10.121 -data ArgusCamera
  

  • Look at the unit_tests/params_test to check for sample usage of parameters in test code

Ubuntu 18.04 and 20.04 x64

Prerequisites

• Run the following to get latest build tools

  sudo apt-get update && sudo apt-get -y install   autoconf   automake  autopoint  build-essential  git-core  git-lfs libass-dev   libfreetype6-dev  libgnutls28-dev   libmp3lame-dev libsdl2-dev  libtool libsoup-gnome2.4-dev libncurses5-dev libva-dev   libvdpau-dev   libvorbis-dev   libxcb1-dev   libxcb-shm0-dev   libxcb-xfixes0-dev  ninja-build   pkg-config   texinfo   wget   yasm   zlib1g-dev   nasm   gperf bison curl zip unzip tar python3-pip flex && pip3 install meson
  

• Note: start a new terminal as pip3 settings do not get effective on the same shell
• CMake minimum version 3.22 - Follow this article to update cmake
• Clone with submodules and LFS.

  git clone --recursive https://github.com/Apra-Labs/ApraPipes.git
  

• Run ./bootstrap-vcpkg.sh in vcpkg/ directory
• Run ./vcpkg integrate install
• build gstreamer
  • cd thirdparty && sh ./build_gstreamer.sh && cd -
  • update .bashrc and append following line at the end of it. Adjust the path based on your environment.

    export LD_LIBRARY_PATH=~/ApraPipes/thirdparty/gst-build/gst-build-1.16/outInstall/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH
    

  • load symbols from .bashrc source ~/.bashrc.

Build for linux

• chmod +x build_linux_x64.sh or chmod +x build_linux_no_cuda.sh
• ./build_linux_x64.sh or ./build_linux_no_cuda.sh depending on previous step. No Cuda as the name suggests will not build the Nvidia Cuda GPU Modules

Build can take ~2 hours depending on the machine configuration.

Build and test using docker

• Use this docker image with all the software setup.

  docker pull ghcr.io/kumaakh/aprapipes-build-x86-ubutu18.04-cuda:latest
  

• Run the docker container using above image
• Mount an external volume as a build area
• clone the repository with submodules and LFS as described above
• build using build_linux_*.sh scripts as described above

This build will be fairly fast (~10 mins) as entire vcpkg cache comes down with the docker image

Jetson boards - Nano, TX2, NX, AGX

Prerequisites

• Setup the board with Jetpack 4.4
• run the following

  sudo apt-get update && sudo apt-get -y install git-lfs libncurses5-dev ninja-build nasm curl libudev-dev && sudo snap install cmake --classic
  

• append following lines to ~/.bashrc

  export VCPKG_FORCE_SYSTEM_BINARIES=1
  export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}
  export LD_LIBRARY_PATH=/usr/local/cuda/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
  

• reload ~/.bashrc:

  source ~/.bashrc:
  

• Clone with submodules and LFS.

  git clone --recursive https://github.com/Apra-Labs/ApraPipes.git
  

• Run ./bootstrap-vcpkg.sh in vcpkg/ directory
• Run ./vcpkg integrate install

Build for jetsons

• chmod +x build_jetson.sh
• ./build_jetson.sh

Build can take ~12 hours on Jetson Nano. Note: Jetson build can also be done using Ubuntu 18.04 x86_64 Laptop via cross compilation.

Cross compilation using qemu

Conceptual steps adapted from here:

• On any Intel Ubuntu 18.04 computer (physical or virtual including wsl ) mount a Jetson SD Card Image as described above
• Copy relevant files from mounted image to created a rootfs
• Install qemu on ubuntu host
• chroot into emulated aarm64 environment using script provided in the github link above
• install extra tools and build aprapipes and aprapipesut
• the built aprapipesut can be copied to a Jetson board and run.

This approach can use all 12-16 cores of a laptop and hence builds faster.

Run Tests

• list all tests _build/aprapipesut --list_content
• run all tests _build/aprapipesut
• run one test _build/aprapipesut --run_test=filenamestrategy_tests/boostdirectorystrategy
• run one test with arguments _build/aprapipesut --run_test=unit_tests/params_test -- -ip 10.102.10.121 -data ArgusCamera
  • Look at the unit_tests/params_test to check for sample usage of parameters in test code

This project uses boost tests for unit tests.

Update Submodules

git submodule update --init --recursive

Update Documentation

If any changes are made in the documentation i.e. in /docs/source folder, the docs must be regenerated again follwing the steps given below. New contents from the /docs/build directory should be committed.

To regenerate documentation

To build docs
apt-install get python-sphinx 
pip install sphinx-rtd-theme
cd docs
make html