Installation

This work flow has been successfully tested and run on the following systems:

-Intel Core i5-2400 3.10Ghz Quad Core, 8 GB RAM, nVidia  GTX 460 V2 Fermi 1GB 336 processing cores
-Intel Pentium G6950 2.8Ghz Dual Core, 16 GB RAM, nVidia Tesla C1060 4GB 240 processing cores
-(Dual) Intel Xeon X5680 3.33Ghz Hex Core Hyper-threaded, 72GB RAM, nVidia GT430 2GB 96 processing cores

Requirements

This work flow is quite computationally expensive when dealing with large image sets, therefore the system that it should be run on should be a 64-bit system with at least 4 GB of RAM.
This work flow also uses a GPU accelerated version of SIFT, so the system also needs a CUDA-capable NVIDIA GPU.
- The libraries which siftGPU use can be run with both CUDA and OpenGL-based GPUs however an OpenGL-based GPU has not been tested with this work flow.

The following shows the steps for installing this work flow on a fresh install of Fedora 16 64-bit, with a CUDA-capable GPU. It is assumed that the user has root access.

Installation Steps

A number of dependencies are used in this work flow, the following can be installed using yum

Update

$ sudo yum update -y --skip-broken

Reboot if needed then perform the the following command

$ sudo yum install git patch make gcc-c++ freeglut-devel libXi-devel
libXmu-devel DevIL-devel boost-devel gsl-devel libjpeg-devel
lapack-devel zlib-devel opencv opencv-devel

The proper NVIDIA drivers and CUDA have to be installed, this process can vary across systems. The following steps (taken from here) were found to work with a standard kernel.
Install the RPM fusion repositories

$ sudo yum localinstall --nogpgcheck
http://download1.rpmfusion.org/free/fedora/rpmfusion-free-release-stable.noarch.rpm
http://download1.rpmfusion.org/nonfree/fedora/rpmfusion-nonfree-release-stable.noarch.rpm

Install the NVIDIA drivers

$ sudo yum install kmod-nvidia akmod-nvidia kernel-devel

Back up the initramfs image and generate a new one

$ sudo mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname-r).nouveau.img
$ sudo dracut /boot/initramfs-$(uname -r).img $(uname -r)

Run nvidia-xconfig

$ sudo nvidia-xconfig

Reboot and then download the Cuda Tool Kit for Fedora
Make the CUDA tool kit executable and install it as root.
- NOTE: If you install CUDA to a place other than the default (/usr/local/) keep track of this, you will need it. You will also need to change all /usr/local/cuda paths in the rest of the installation process to your chosen path

$ chmod +x cudatoolkit
$ sudo cudatoolkit

Edit line 80 of the CUDA host_config file to allow for gcc 4.6 to be used

In /usr/local/cuda/include/host_config.h

Change Line 80 to:

#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 7)

Add the following to your ~/.bashrc file then reload the shell to update the environment variables.

In ~/.bashrc

export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib:/usr/local/cuda/lib64

GLEW has to be built from source for siftGPU. Clone the source into a directory of your choice. Then build and install it.

$ git clone git://glew.git.sourceforge.net/gitroot/glew/glew
$ cd glew/auto/
$ make
$ cd ../
$ sudo make install

Download siftGPU from the site, then go into the siftGPU directory and run make.
- NOTE: If you installed cuda to anywhere besides /usr/local you will have to edit line 11 of the makefile to reflect that.

$ cd SiftGPU
$ make

Then move the compiled library to /usr/lib and copy the headers to /usr/include

$ sudo mv bin/libsiftgpu.so /usr/lib
$ sudo cp -p src/SiftGPU/*.h /usr/include

Next version 1.2 of Graclus needs to be downloaded for CMVS. Once downloaded, edit the Makefile.in varible DNUMBITS to equal 64 instead of 32

In graclus1.2/Makefile.in

Change line 11 to:

COPTIONS = -DNUMBITS=64

Run make in the graclus directory, move the compiled libraries to /usr/lib and the metis header files to /usr/include

$ cd graclus1.2
$ make
$ sudo mv *.a /usr/lib
$ sudo cp -p metisLib/*.h /usr/include

Clone our 3d-work flow into a working directory and run the setup for the source tree. This set up will download the source files for Bundler, PMVS2, and CMVS from their respective websites. It will then apply the provided patches to the source code in order to replicate the source tree that we use.

$ git clone https://github.com/drn2369/workflow-3d.git
$ cd workflow-3d/src
$ sh setup.sh

Run make and the binaries should compile and be placed in ../bin

$ make

Move libANN_char.so to /usr/lib and run ldconfig to update the links

sudo mv bundler-v0.4-source/lib/libANN_char.so /usr/lib
sudo ldconfig

Finally, update the path in the RunProcess script in workflow-3d/scripts/ so the base path is the full path to workflow-3d

In RunProcess.sh:

BASEPATH=/full/path/to/workflow-3d

BINPATH=$BASEPATH/bin
SCRIPTPATH=$BASEPATH/scripts

SIFT=$BINPATH/siftGPU
MATCHKEYS=$BINPATH/KeyMatchFull
BUNDLER=$BINPATH/bundler
BUNDLE2PMVS=$BINPATH/Bundle2PMVS
CMVS=$BINPATH/cmvs
GENOP=$BINPATH/genOption

Check the User Guide on how to run the work flow on the provided test data.