PopSift is a GPU implementation that requires an NVIDIA GPU card with a CUDA compute capability >= 3.0 (including, e.g. the GT 650M). The code is originally developed with the compute capability 5.2 card GTX 980 Ti in mind.
Here are the minimum hardware requirements for PopSift:
Windows x64, Linux, macOS
Recent Intel or AMD cpus
No particular requirements
NVIDIA CUDA-enabled GPU (compute capability >= 3.5)
The core library depends only on Cuda >= 7.0
The library includes a few sample applications that show how to use the library. They require
Boost >= 1.55 (required components atomic, chrono, date-time, system, thread)
[optionally] DevIL (libdevil-dev) can be used to load a broader range of image formats, otherwise only pgm is supported.
vcpkg is a cross-platform (Windows, Linux and MacOS), open-source package manager created by Microsoft.
Starting from v0.9, PopSift package can be installed on each platform via vcpkg. To install the library:
vcpkg install popsift --triplet <arch>
<arch> is the architecture to build for (e.g.
If you want to install the demo applications that come with the library you can add the option
vcpkg install popsift[apps] --triplet <arch>
Building the library¶
CMake >= 3.14 to build the code
C/C++ compiler supporting the C++11 standard (gcc >= 4.6 or visual studio or clang)
CUDA >= 7.0
vcpkg can be used to install all the dependencies on all the supported platforms. This is particularly useful on Windows. To install the dependencies:
vcpkg install cuda devil boost-system boost-program-options boost-thread boost-filesystem
You can add the flag
--triplet to specify the architecture and the version you want to build.
--triplet x64-windowswill build the dynamic version for Windows 64 bit
--triplet x64-windows-staticwill build the static version for Windows 64 bit
--triplet x64-linux-dynamicwill build the dynamic version for Linux 64 bit
and so on. More information can be found here
On Linux you can install from the package manager:
For Ubuntu/Debian package system:
sudo apt-get install g++ git-all libboost-all-dev libdevil-dev
For CentOS package system:
sudo yum install gcc-c++ git boost-devel devil
On MacOs using Homebrew install the following packages:
brew install git boost devil
Getting the sources¶
git clone https://github.com/alicevision/PopSift.git
From PopSift root folder you can run cmake:
mkdir build && cd build cmake .. make -j `nproc`
On Windows add
-G "Visual Studio 16 2019" -A x64 to generate the Visual Studio solution according to your VS version (see CMake documentation).
If you are using the dependencies built with VCPKG you need to pass
-DCMAKE_TOOLCHAIN_FILE=path/to/vcpkg/scripts/buildsystems/vcpkg.cmake at cmake step to let it know where to find the dependencies.
CMake configuration can be controlled by changing the values of the following variables (here with their default value)
BUILD_SHARED_LIBS:BOOL=ONto enable/disable the building shared libraries
PopSift_BUILD_EXAMPLES:BOOL=ONto enable/disable the building of applications
PopSift_BUILD_DOC:BOOL=OFFto enable/disable building this documentation and the Doxygen one.
For example, if you do not want to build the applications, you have to pass
-DPopSift_BUILD_EXAMPLES:BOOL=OFF and so on.
PopSift as third party¶
When you install PopSift a file
PopSiftConfig.cmake is installed in
<install_prefix>/lib/cmake/PopSift/ that allows you to import the library in your CMake project.
CMakeLists.txt file you can add the dependency in this way:
1# Find the package from the PopSiftConfig.cmake 2# in <prefix>/lib/cmake/PopSift/. Under the namespace PopSift:: 3# it exposes the target PopSift that allows you to compile 4# and link with the library 5find_package(PopSift CONFIG REQUIRED) 6... 7# suppose you want to try it out in a executable 8add_executable(popsiftTest yourfile.cpp) 9# add link to the library 10target_link_libraries(popsiftTest PUBLIC PopSift::PopSift)
Then, in order to build just pass the location of
PopSiftConfig.cmake from the cmake command line:
cmake .. -DPopSift_DIR=<install_prefix>/lib/cmake/PopSift/
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/ )
Building the dependency image¶
We provide a
Dockerfile_deps containing a cuda image with all the necessary PopSift dependencies installed.
CUDA_TAG can be passed when building the image to select the cuda version.
OS_TAG can be passed to select the Ubuntu version.
For example to create the dependency image based on ubuntu 18.04 with cuda 8.0 for development, use
docker build --build-arg CUDA_TAG=8.0 --tag alicevision/popsift-deps:cuda8.0-ubuntu18.04 -f Dockerfile_deps .
The complete list of available tags can be found on the nvidia [dockerhub page](https://hub.docker.com/r/nvidia/cuda/)
Building the PopSift image¶
Once you built the dependency image, you can build the popsift image in the same manner using
docker build --tag alicevision/popsift:cuda8.0-ubuntu18.04 .
Running the PopSift image¶
In order to run the image nvidia docker is needed: see the installation instruction. Once installed, the docker can be run, e.g., in interactive mode with
docker run -it --runtime=nvidia alicevision/popsift:cuda8.0-ubuntu18.04
Official images on DockeHub¶
Check the docker hub PopSift repository for the available images.