Installation Instructions

You can either compile bob or use the provided binary packages.

Binary Installation

We offer pre-compiled binary installations of Bob using conda.

Conda (Linux and MacOSX 64-bit)


Please install and get familiar with conda first by referring to their website before getting started.

The commands below will you give you a minimal bob environment. Continue reading for more detailed instructions.

$ conda update -n root conda
$ conda config --set show_channel_urls True
$ conda config --add channels defaults
$ conda config --add channels
$ conda create -n bob_py27 --override-channels -c -c defaults python=2.7 bob
$ source activate bob_py27
$ python -c 'import'


The packages in our channel are not API/ABI compatible with packages in other user/community channels (especially conda-forge). We can only guarantee that the packages in our channel is compatible with the defaults channel.

Detailed installation instructions

Anaconda is a python distribution which comes with conda as its package manager besides pip. conda can package every program and not just python libraries. conda packages are usually compiled binary packages and you do not need to compile them.

To get started install Miniconda (recommended) or Anaconda (Anaconda comes with Miniconda and some extra packages already installed).

  • Make sure that conda is in your path and get familiar with conda by reading their webpage:
$ which conda
$ conda update -n root conda
  • Add our conda channel:
$ conda config --add channels defaults
$ conda config --add channels
  • Create a new environment:
$ conda create -n bob_py27 python=2.7 bob


you can specify the version of python and bob that you want to use here. For example, conda create -n bob-2.5.1_py35 python=3.5 bob=2.5.1.

  • Activate the environment:
$ source activate bob_py27


You always need to activate your environments using source activate environment_name before you start using them (each time you open a new terminal).

  • Install other conda packages that you may want to use:
$ conda install sphinx nose
  • Bob has many packages but only the core packages get installed with bob (mainly the C++ packages). You can install the extra packages using pip. For example, to install the framework:
$ pip install
  • To see which bob packages are already installed:
$ conda list bob

And now you are ready to run your programs.

>>> import
>>> ...


Now that you have all the dependencies and required Bob packages installed, it is a good idea to follow our Using zc.buildout page to see how you can fine tune your installations. Remember to use the python that you installed with conda create -n ... earlier. Our zc.buildout scripts usually pulls some extra python libraries such as sphinx and nose. You can install those with conda first to avoid their compilation on your pc.

Stable bob installations

You can use the anaconda package to get super stable Bob installations. However, you will end up with very large environments with a lot of packages that you may not use. To create such an environment:

export PY_VER="2.7"
export BOB_VER="2.6.1"
export ANACONDA_VER="4.3.0"
export ENV_NAME="bob-$BOB_VER-$PY_VER"
conda create --yes -n $ENV_NAME --override-channels -c -c defaults python=$PY_VER anaconda=$ANACONDA_VER bob=$BOB_VER
source activate $ENV_NAME

There is only one compatible anaconda version for each version of Bob. For example bob 2.5.1 works only with anaconda 4.2.0 (as far as we have tested). Here is a table that lists compatible versions:

Bob version Anaconda version
2.5.1 4.2.0
2.6.2 4.3.0

Compiling from Source

Following, you will find the software dependencies required for Bob’s compilation and instructions on how to install all the equivalent software packages.


This section describes software dependencies required for Bob’s compilation and runtime.


We keep here a comprehensive list of all packages you may need to compile most of the available Bob packages. You may not need all this software for special deployments. You should choose whatever suits you best. If you have problems or would like to report a success story, please use our mailing list for discussions.

Library Min. Versio n License Notes
Std. C/C++ Libraries any Depends on the compiler Required by all packages with C/C++ bindings
Blitz++ 0.10 Artistic-2.0 or LGPLv3+ or GPLv3+ Required by all packages with C/C++ bindings
Lapack any BSD-style Required by bob.math
Python 2.5 Python-2.0 Required by all packages
Boost 1.34 BSL-1.0 Required by all packages with C/C++ bindings
HDF5 1.8.4 HDF5 License (BSD-like, 5 clauses) Required by all I/O operations (direct or indirect dependencies to
libpng 1.2.42 ? libpng license Required by all packages that do image I/O and manipulation (depend directly or indirectly on
libtiff 3.9.2 BSD-style Required by all packages that do image I/O and manipulation (depend directly or indirectly on
giflib 4.1.6- 9 MIT Required by all packages that do image I/O and manipulation (depend directly or indirectly on
libjpeg 6.2? GPL-2.0 or later (also commercial) Required by all packages that do image I/O and manipulation (depend directly or indirectly on
FFMpeg or libAV 0.5 (ffmpe g) or 0.8 (libav ) LGPL-2.1 or later, or GPL-2.0 or later Required by all packages that do video I/O and manipulation (depend directly or indirectly on
MatIO 1.3.3? BSD-2-Clause Required by all packages that do Matlab I/O and manipulation (depend directly or indirectly on
VLFeat 0.9.14 BSD-2-Clause Required by bob.ip.base and all dependents
LIBSVM 2.89+ BSD-3-Clause Required by bob.learn.libsvm and all dependents
CMake 2.8 BSD-3-Clause Required by all packages with C/C++ bindings. Use at least CMake 2.8.12 on Mac OS X.
Dvipng 1.12? GPL-3.0 Required by all packages (documentation generation)
LaTeX any ? Required by all packages (documentation generation). You will also need to install the Extra-Latex fonts package.
pkg-config any GPL-2.0 Required to find dependencies while building bob packages.

Here is a list of Python packages software that is also used by Bob. It is not required that such software be installed at the moment you compile Bob. It will be fetched automatically from PyPI otherwise.

Library Min . Ver sio n License Notes
NumPy 1.3 BSD-3-Clause Required by all packages. If not installed natively on your machine, may not correctly use optimized LaPACK or BLAS implementations. Consequently, bob.math will not either.
SciPy 0.7 ? BSD-3-Clause Required at least by bob.ap, bob.learn.boosting, bob.ip.optflow.hornschunk and facereclib
Matplotlib 0.9 9 Based on Python-2.0 Required for plotting
SQLAlchemy 0.5 MIT Required by all database accessor packages (i.e., any that starts with bob.db)
nose 1.0 ? LGPL-2.1 For unit testing, on all packages
Sphinx 0.6 BSD-2-Clause Required by all packages (documentation generation)
Setuptools 8.0 Python-2.0 Required by all packages (Buildout and package compilation)
Pillow 1.7 .x? BSD-like Required by at least and bob.ip.optflow.liu
IPython any BSD-3-Clause Recommended as interactive prompt

Setting up a Development Environment using Conda

Please look at this script that lets you setup a conda environment for Bob development and installs all the dependencies in one shot.

Installing Bob from source

Once the dependecies are installed you can use buildout (for production) or pip (for experts) to install Bob from source.

Using zc.buildout (for production)

Please read our Using zc.buildout page for instructions on how to install Bob packages. This is the recommended way to install bob from PyPI instead of pip.

Using pip (for experts)

It is also possible to install Bob packages using pip, globally or on your private virtualenv, if that is the way you like your Python work environments. Differently than with zc.buildout, you will need to manually install all packages you need (directly or indirectly), as pip/setuptools has presently no way to coherently install Python packages that depend on each other for building, such as is the case of many Bob packages.

For example, to install in a newly created virtual environment, here is the sequence of commands to execute:

$ pip install numpy
$ pip install bob.extension
$ pip install bob.blitz
$ pip install bob.core
$ pip install
$ pip install


Each pip command must be executed separately, respecting the inter- package dependency.

The following will not work as expected:

$ #Do not do this:
$ pip install numpy

The dependency of Bob core packages can be summarized into 8 layers and the following script can be used to install all core Bob packages using pip:

$ bash
set -e

get_layer() {
case $1 in
    packages=("bob.core" "bob.ip.draw")
    packages=("" "bob.sp" "bob.math")
    packages=("bob.ap" "bob.measure" "bob.db.base" "" "" "" "bob.ip.base" "bob.ip.color" "bob.ip.gabor" "bob.learn.activation" "bob.learn.libsvm" "bob.learn.boosting")
    packages=("" "bob.learn.linear" "bob.learn.mlp" "" "bob.db.mnist" "bob.db.atnt" "bob.ip.flandmark" "bob.ip.facedetect" "bob.ip.optflow.hornschunck" "bob.ip.optflow.liu")
    packages=("bob.learn.em" "bob.db.iris")

for layer in `seq 1 8`;
  get_layer ${layer}
  for pkg in "${packages[@]}";
    pip install $pkg
Hooking-in privately compiled externals

If you have placed external libraries outside default search paths, make sure you set the environment variable BOB_PREFIX_PATH to point to the root of the installation paths for those, before you run pip install...:

$ export BOB_PREFIX_PATH="/path/to/my-install:/path/to/my-other-install"
$ pip install numpy
$ pip install
Package development

You can develop your package as per-usual, no special restrictions apply. You may refer to our Bob Satellite Package Development and Maintenance for further hints and tips. In this case, ignore zc.buildout setup instructions.

Available packages

For a comprehensive list of packages that are either part of Bob or use Bob, please visit packages.

Supported platforms

Currently our packages are being compiled and tested against Python 2.7, 3.4, and 3.5 on Linux and MacOSX 64bit machines.