Skip to content

Latest commit

 

History

History
99 lines (71 loc) · 3.95 KB

INSTALL.md

File metadata and controls

99 lines (71 loc) · 3.95 KB
Raw

CCTag library

Building instructions

Required tools:

Optional tool:

  • CUDA >= 9.0 Note: On Windows, there are compatibility issues to build the GPU part due to conflicts between msvc/nvcc/thrust/eigen/boost.

Getting the sources:

$ git clone https://github.com/alicevision/CCTag.git

Dependencies

Most of the dependencies can be installed from the common repositories (apt, yum etc):

  • Eigen3 (libeigen3-dev) >= 3.3.4 (NOTE: in order to have Cuda support on Windows, at least version 3.3.9 is required)
  • Boost >= 1.66 ([accumulators, atomic, chrono, core, date-time, exception, filesystem, math, program-options, ptr-container, system, serialization, stacktrace, timer, thread]-dev)
  • OpenCV >= 3.1
  • TBB >= 4.0

On a recent Ubuntu-like distribution (e.g. 14.04), you may want to try to run:

$ sudo apt-get install g++ git-all libpng12-dev libjpeg-dev libeigen3-dev libboost-atomic-dev libboost-chrono-dev libboost-date-time-dev libboost-dev libboost-program-options-dev libboost-exception-dev libboost-filesystem-dev libboost-serialization-dev libboost-system-dev libboost-thread-dev libboost-timer-dev libtbb-dev

OpenCV need to be compiled separately and installed in some OPENCV_INSTALL path. Then, when running cmake you need to provide the path to the location where OpenCVConfig.cmake is installed, usually ${OPENCV_INSTALL}/share/share/OpenCV/ (see below).

CCTag contains code optimized for AVX2 instruction set, which significantly increases detection performance. You can enable it with the option: cmake -DCCTAG_ENABLE_SIMD_AVX2=ON.

Run the building process

You now just need to be in the CCTag folder and run cmake:

$ mkdir build && cd build
$ cmake .. -DOpenCV_DIR=${OPENCV_INSTALL}/share/share/OpenCV/
$ make -j `nproc`

Use -DCCTAG_WITH_CUDA:BOOL=ON to enable (default) or disable the cuda support. On Windows, it must be set to OFF if the Eigen version is < 3.3.9

If you want to install the library to, say, a CCTAG_INSTALL path, just add -DCMAKE_INSTALL_PREFIX=$CCTAG_INSTALL at cmake command line. If you want to build CCTag as a shared library: -DBUILD_SHARED_LIBS=ON.

Using CCTag as third party

When you install CCTag a file CCTagConfig.cmake is installed in $CCTAG_INSTALL/lib/cmake/CCTag/ that allows you to import the library in your CMake project. In your CMakeLists.txt file you can add the dependency in this way:

# Find the package from the CCTagConfig.cmake
# in <prefix>/lib/cmake/CCTag/. Under the namespace CCTag::
# it exposes the target CCTag that allows you to compile
# and link with the library
find_package(CCTag CONFIG REQUIRED)
...
# suppose you want to try it out in a executable
add_executable(cctagtest yourfile.cpp)
# add link to the library
target_link_libraries(cctagtest PUBLIC CCTag::CCTag)

Then, in order to build just pass the location of CCTagConfig.cmake from the cmake command line:

cmake .. -DCCTag_DIR=$CCTAG_INSTALL/lib/cmake/CCTag/

Docker Image

A docker image can be built using the Ubuntu based Dockerfile,which is based on nvidia/cuda image (https://hub.docker.com/r/nvidia/cuda/)

A parameter CUDA_TAG can be passed when building the image to select the ubuntu and cuda version. For example to create a ubuntu 16.04 with cuda 8.0 for development, use

docker build --build-arg CUDA_TAG=8.0-devel --tag cctag .

The complete list of available tags can be found on the nvidia dockerhub page In order to run the image nvidia docker is needed: see the installation instruction here https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0) Once installed, the docker can be run, e.g., in interactive mode with

docker run -it --runtime=nvidia cctag