FIREUbuntu104

Step-by-step instruction to compile FIRE on Ubuntu 10.4.

To facilitate installation, I provide here a description how to compile the FIRE server on Ubuntu 10.4 from scratch.

The description applies to FIRE V 2.3 (SVN checkout of July 26, 2010).

Compiling FIRE

• default Ubuntu 10.4 installation (in May 2010 in virtualbox)
• upgrade to recent state (sudo apt-get update; sudo apt-get dist-upgrade) on July 26, 2010
• sudo apt-get install subversion g++ libmagick++-dev
• svn checkout http://fire-cbir.googlecode.com/svn/trunk/ fire-cbir
• cd fire-cbir
• make

After that the bin-Directory contain a total of 47 binaries and the FIRE server is ready to use.

Integration with Apache

• sudo apt-get install apache2
• cd ~/fire-cbir/
• sudo cp Python/firesocket.py /usr/lib/cgi-bin
• cd WebInterface
• sudo cp fire.py feature.py img.py config.py fire-template.html /usr/lib/cgi-bin
• sudo mkdir /usr/lib/cgi-bin/images
• sudo cp neutral.png positive.png negative.png fire-logo.png i6.png /usr/lib/cgi-bin/images

Now all the files should be in place for the FIRE webinterface to work

Setting up a dataset

• create a directory mkdir ~/fire-img
• put a set of medium sized images (e.g. 500x500 pixels) into this directory. For a start use about 100 images.
• convert all images into jpg (not necessary but often avoids problems)
• cd ~/fire-img
• mogrify -format jpg *
• extract color histograms
• ~/fire-cbir/bin/extractcolorhistograms --color --images ~/fire-img/*.jpg
• create a filelist for this dataset
• cd ~/fire-img/
• echo FIRE_filelist > filelist
• echo suffix color.histo.gz >> filelist
• echo path this >> filelist
• ls *.jpg | sed 's/^/file /' >> filelist

Now you have a dataset where every image is represented using a color histogram. Other features can be added by extracting them and adding the corresponding suffix lines to the filelist.

Starting the FIRE server and accessing it through the webinterface

• cd ~/fire-cbir
• ./bin/fire -f ~/fire-img/filelist -r 10
• Access the fire webinterface throught http://localhost/cgi-bin/fire.py
• There you should be able to click an image and get similar images (according to color histograms)