Installation¶

Our input for the competition uses the signal-processing and machine learning toolbox called Bob. Follow bellow the instructions to, first, install Bob and then install our competition software.

1 - Installing bob¶

Bob is a signal-processing and machine learning toolbox originally developed by the Biometrics Group at Idiap, in Switzerland.

We offer pre-compiled binary installations of Bob (v2) using conda for Linux and MacOSX. The sequence of commands below will get you started and will install all required dependencies.

# for linux
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh

# for mac
wget https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh -O ~/miniconda.sh

bash ~/miniconda.sh
# Make sure you follow instructions and modify your
# $PATH variable according to the installation instructions.
mv ~/.condarc ~/.condarc.bak
# remove any conda configuration that might interfere.
# We don't use conda-forge for distributing bob anymore and bob packages
# are not compatible with conda-forge packages. So don't mix them!
conda update -n root conda

# Adding some the defaults and the bob.conda channels
conda config --set show_channel_urls True
conda config --add channels defaults
conda config --add channels https://www.idiap.ch/software/bob/conda

# Creating the conda env
conda create -n bob_env_py27 python=2.7 gcc=4
source activate bob_env_py27

#Install depedendencies
conda install scikit-image
conda install -c bob

# A small test to see if things are working
python -c 'import bob.io.base'

2 - Competition software installation¶

Once Bob is installed, you can install our competition input either via pip or via zc.buildout.

The installation via zc.buildout is easier to debug and since this submission is intermediate, we will describe only the installation via this zc.buildout. Follow bellow the steps to install it.

# Be sure that you have the conda-env that you created in the last step activated
source activate bob_env_py27

# Cloning the repo
git clone https://gitlab.idiap.ch/bob/bob.bio.pericrosseye_competition/
cd bob.bio.pericrosseye_competition

# Checking out the last tag
git checkout -b v1.0.2 tags/v1.0.2

# Installation
python bootstrap-buildout.py
./bin/buildout

# Most important step. THE UNIT TESTS
./bin/nosetests -sv bob.bio.pericrosseye_competition

Competition scripts¶

This section describes how to run the competition scripts. Implementation and fine tunning details are described here.

We provide two periocular recognition systems. The first one is based on Intersession Variability Modelling [FRE16] and the second one is based on Geodesic Flow Kernel with gabor jets.

Intersession Variability Modelling¶

Description¶

Built on top of Gaussian Mixture Models (GMM), Intersession Variability Modelling (ISV) proposes to explicitly model the variations between different modalities by learning a linear subspace in the GMM supervector space. These variations are compensated during the enrolment and testing time. For this input, the periocular images are resized to 90 x 90 pixels and sampled in patches of 12 × 12 pixels moving the sampled window in one pixel. Then each patch is mean and variance normalized and the first 45 DCT coefficients are extracted. The Universe Background Model ( $UBM$ ) is modelled with 512 gaussians and the dimension of the session varibility matrix ( $U$ ) is 160. Implementation details of this input can be found in [FRE16].

Scripts¶

The script perienroll_idiap.py, enrolls a client given an image using this system. Follow bellow the help message of the script.

./bin/perienroll_idiap.py --help

Enroll a client using Intersession Variability Modelling

Usage:
  perienroll_idiap.py <image_file> <template_file> <output_file> [-v]
  perienroll_idiap.py -h | --help
Options:
  -h --help           Show this screen.
  -v                  Verbosity level

Warning

The <template_file> argument must have the extension .hdf5 (example: template_client_001.hdf5).

The scoring script is carryed out using the script perimatch_idiap.py.

./bin/perimatch_idiap.py --help
Do the matching using Intersession Variability Modelling
Usage:
  perimatch_idiap.py <image_file> <template_file> <output_file> [-v]
  perimatch_idiap.py -h | --help
Options:
  -h --help           Show this screen.
  -v                  Verbosity level

Warning

The <template_file> argument must have the extension .hdf5 (example: template_client_001.hdf5).

Geodesic Flow Kernel on Gabor Graphs¶

Description¶

The Geodesic Flow Kernel (GFK) models the source domain and the target domain with d-dimensional linear subspaces and embeds them onto a Grassmann manifold. Then a Geodesic Flow [Gong12] between these two subspaces ( $G$ ) is built and an infinite number of subspaces is integrated along the flow. A grid of gabor jets along the periocular image are as features. A comparison between two grids of gabor jets from visible light and near infra-red respectivelly $S_n$ and $T_n$ can be done as: $\frac{\sum_{n=1}^{N} S_n \cdotp G \cdotp T_n}{N}$ .

Scripts first setup¶

The script perienroll_idiap_gfk.py, enrolls a client given an image using this system. Follow bellow the help message of the script.

./bin/perienroll_idiap_gfk.py

Enroll a client using the Gabor Graph with Geodesic Flow Kernel

Usage:
  perienroll_idiap_gfk.py <template_file_A> <template_file_B> <output_file> [-v]
  perienroll_idiap_gfk.py -h | --help
Options:
  -h --help           Show this screen.
  -v                  Verbosity level

Warning

The <template_file> argument must have the extension .hdf5 (example: template_client_001.hdf5).

The scoring script is carryed out using the script perimatch_idiap_gfk.py.

./bin/perimatch_idiap_gfk.py --help

Do the matching using the Gabor Graph with Geodesic Flow Kernel

Usage:
  perimatch_idiap_gfk.py <template_file_A> <template_file_B> <output_file> [-v]
  perimatch_idiap_gfk.py -h | --help
Options:
  -h --help           Show this screen.
  -v                  Verbosity level

Warning

The <template_file> argument must have the extension .hdf5 (example: template_client_001.hdf5).

Scripts second setup¶

This system is the same as the one above, but with a different setup.

The script perienroll_idiap_gfk_10d.py, enrolls a client given an image using this system. Follow bellow the help message of the script.

./bin/perienroll_idiap_gfk.py

Enroll a client using the Gabor Graph with Geodesic Flow Kernel

Usage:
  perienroll_idiap_gfk_10d.py <template_file_A> <template_file_B> <output_file> [-v]
  perienroll_idiap_gfk_10d.py -h | --help
Options:
  -h --help           Show this screen.
  -v                  Verbosity level

Warning

The <template_file> argument must have the extension .hdf5 (example: template_client_001.hdf5).

The scoring script is carryed out using the script perimatch_idiap_gfk_10d.py.

./bin/perimatch_idiap_gfk_10d.py --help

Do the matching using the Gabor Graph with Geodesic Flow Kernel

Usage:
  perimatch_idiap_gfk_10d.py <template_file_A> <template_file_B> <output_file> [-v]
  perimatch_idiap_gfk_10d.py -h | --help
Options:
  -h --help           Show this screen.
  -v                  Verbosity level

Warning

The <template_file> argument must have the extension .hdf5 (example: template_client_001.hdf5).