SiftGPU -- Compile and Use

Requirements

The SiftGPU is compiled on Linux OS, which requires these environments and tools:

• Ubuntu 16.04 LTS
• cmake 3.5 or cmake 2.8
• OpenCV2.4.13
• Glew 1.13.0
• OpenGL
• DevilImage
• Others

Compile

Install OpenGL

sudo apt-get install libgl1-mesa-dev libglu1-mesa-dev freeglut3-dev libdevil-dev

Download SiftGPU

git clone https://github.com/pitzer/SiftGPU.git
cd SiftGPU

Or you can download the project on author's page. But this link sames to be lost.

Download and install Glew

Download Glew from here. Please choose the version of 1.13.0, whatever .zip or .tgz file format, extract it and cd into that path, then

sudo make
sudo make install

This will produce a lib file named libGLEW.so.1.13 in the directory lib, and then

sudo cp lib/libGLEW.so.1.13 /usr/lib64/
sudo ldconfig /usr/lib64/

Compile SiftGPU

You can compile SiftGPU in two ways: Without CUDA acceleration i.e. CPU-Only and using CUDA, i.e. using GPU.

• CPU-Only

cd SiftGPU
sudo make

• Using GPU

cd SiftGPU
sudo vim Makefile
# 1. modify the siftgpu_enable_cuda=0 to siftgpu_enable_cuda=1 on two positions
# 2. modify the CUDA_INSTALL_PATH to where CUDA is installed, e.g.'/usr/local/cuda-9.0'
# 3. In addition, you can choose to change the siftgpu_cuda_options in your specific setting
# Finally, save and exit.
make

There should be a Lib file named libsiftgpu.so in directory bin, you can easily check this Lib file whether it is correct.

ldd bin/libsiftgpu.so

which should output the linking information just like these:

linux-vdso.so.1 =>  (0x00007ffe2c182000)
    libGLEW.so.1.10 => /usr/lib/x86_64-linux-gnu/libGLEW.so.1.10 (0x00007ff2c899f000)
    libglut.so.3 => /usr/lib/x86_64-linux-gnu/libglut.so.3 (0x00007ff2c8755000)
    libGL.so.1 => /usr/lib/nvidia-384/libGL.so.1 (0x00007ff2c84b1000)
    libIL.so.1 => /usr/lib/x86_64-linux-gnu/libIL.so.1 (0x00007ff2c8196000)
    libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007ff2c7e83000)
    libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007ff2c7b7d000)
    libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007ff2c7966000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007ff2c759d000)
    libX11.so.6 => /usr/lib/x86_64-linux-gnu/libX11.so.6 (0x00007ff2c7268000)
    libXi.so.6 => /usr/lib/x86_64-linux-gnu/libXi.so.6 (0x00007ff2c7058000)
    libXxf86vm.so.1 => /usr/lib/x86_64-linux-gnu/libXxf86vm.so.1 (0x00007ff2c6e52000)
    libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007ff2c6c4e000)
    libGLX.so.0 => /usr/lib/nvidia-384/libGLX.so.0 (0x00007ff2c6a1e000)
    libGLdispatch.so.0 => /usr/lib/nvidia-384/libGLdispatch.so.0 (0x00007ff2c6750000)
    libtiff.so.5 => /usr/lib/x86_64-linux-gnu/libtiff.so.5 (0x00007ff2c64dd000)
    libpng12.so.0 => /usr/lib/x86_64-linux-gnu/libpng12.so.0 (0x00007ff2c62b7000)
    libmng.so.2 => /usr/lib/x86_64-linux-gnu/libmng.so.2 (0x00007ff2c603b000)
    liblcms.so.1 => /usr/lib/x86_64-linux-gnu/liblcms.so.1 (0x00007ff2c5e04000)
    libjpeg.so.8 => /usr/lib/x86_64-linux-gnu/libjpeg.so.8 (0x00007ff2c5baf000)
    /lib64/ld-linux-x86-64.so.2 (0x00007ff2c8e88000)
    libxcb.so.1 => /usr/lib/x86_64-linux-gnu/libxcb.so.1 (0x00007ff2c5990000)
    libXext.so.6 => /usr/lib/x86_64-linux-gnu/libXext.so.6 (0x00007ff2c577e000)
    liblzma.so.5 => /lib/x86_64-linux-gnu/liblzma.so.5 (0x00007ff2c555c000)
    libjbig.so.0 => /usr/lib/x86_64-linux-gnu/libjbig.so.0 (0x00007ff2c534e000)
    libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00007ff2c5135000)
    liblcms2.so.2 => /usr/lib/x86_64-linux-gnu/liblcms2.so.2 (0x00007ff2c4ee0000)
    libXau.so.6 => /usr/lib/x86_64-linux-gnu/libXau.so.6 (0x00007ff2c4cdc000)
    libXdmcp.so.6 => /usr/lib/x86_64-linux-gnu/libXdmcp.so.6 (0x00007ff2c4ad6000)

Congratulations, the compile process has done!

Usage

We provide an example using compiled SiftGPU, which you can find it at test_siftgpu.

If you want to compile this example by yourself, you should

cd test_siftgpu
rm -rf build
cp ./FindGlew.cmake CMAKE_PATH
mkdir build && cd build
# Please change the path of the image for test in main.cpp
# save and exit
# Then, change:
#     include_directories("/home/mac-ar/SiftGPU/src/SiftGPU/" ${OpenGL_INCLUDE_DIR} ${CUDA_INCLUDE_DIR})
#     set(SIFTGPU_LIBS "/home/mac-ar/SiftGPU/bin/libsiftgpu.so")
# in test_siftgpu/CMakeLists.txt to satisfy your system setting.
# save and exit
cmake ..
make

CMAKE_PATH is the path where cmake is installed. Usually like /usr/share/cmake-xx/Modules/, or /usr/local/share/cmake-xx/Modules/, please find the correct path of cmake in your own OS. This script guide the cmake to find GLEW. Just ignore the warnings or note information, no errors means successful compilation. An executable file, named testSIFTGPU, is created in build, just run this program to test the performance

./testSIFTGPU

Then the results will be output, you can change the main.cpp for your specific purpose, which could be better than mine.

Solutions

• When you compile, you may face an error like:

CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
GLUT_Xmu_LIBRARY (ADVANCED)
    linked by target "openglsupport" in directory {$PATH}

You can solve this problem by:

sudo apt-get install libxmu-dev

• Running ./testSIFTGPU, you may face this:

freeglut ERROR: Function <glutDestroyWindow> called without first calling 'glutInit'.

You should open src/SiftGPU/LiteWindow.h and find virtual ~LiteWindow() { if(glut_id > 0) glutDestroyWindow(glut_id); }, modify it as below:

virtual ~LiteWindow()   
{  
    if(glut_id > 0) 
    {
    　　int argc = 0;
    　　char** argv;
    　　glutInit(&argc, argv); 
    　　glutDestroyWindow(glut_id); 
    } 
}

• Another error:

error: declaration of ‘operator new’ as non-function SIFTGPU_EXPORT void* operator new (size_t size);

Please open src/SiftGPU/SiftGPU.h, add a head file #include <stddef.h> at the beginning of it. Last but not least, you should make sure that your OpenGL has been installed and DO NOT run this program using Proxy terminal software without OpenGL support. Just run it using your PC or laptop.