build details

Show: section status errors & todos local changes recent changes last change in-page changes feedback controls

How to install Caffe and Tensorflow on the Duckiebot

Modified 2019-09-22 by Andrea Censi

Caffe and TensorFlow are popular deep learning libraries, and are supported by the Intel Neural Computing Stick (NCS).


Modified 2019-09-22 by Andrea Censi

Step 1: install dependencies and clone repository

Modified 2019-09-22 by Andrea Censi

Install some of the dependencies first. The last command “sudo pip install .....” will cause some time.

sudo apt-get install -y gfortran cython
sudo apt-get install -y libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler git
sudo apt-get install --no-install-recommends libboost-all-dev
sudo apt-get install -y python-dev libgflags-dev libgoogle-glog-dev liblmdb-dev libatlas-base-dev python-skimage
sudo pip install pyzmq jsonschema pillow numpy scipy ipython jupyter pyyaml

Then, you need to clone the repo of caffe

git clone

Step 2: compile Caffe

Modified 2019-09-22 by Andrea Censi

Before compile Caffe, you have to modify Makefile.config

cd caffe
cp Makefile.config.example Makefile.config
sudo vim Makefile.config

Then, change four lines from

'#'CPU_ONLY := 1
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib


INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/arm-linux-gnueabihf/hdf5/serial/

Next, you can start to compile caffe

make all
make test
make runtest
make pycaffe

If you did’t get any error above, congratulation on your success. Finally, please export pythonpath

sudo vim ~/.bashrc
export PYTHONPATH=/home/"$USER"/caffe/python:$PYTHONPATH

Step 3: try it out

Modified 2019-09-22 by Andrea Censi

Now, we can confirm whether the installation is successful. Download AlexNet and run caffe time

cd ~/caffe/
python scripts/ models/bvlc_alexnet
./build/tools/caffe time -model models/bvlc_alexnet/deploy.prototxt -weights  models/bvlc_alexnet/bvlc_alexnet.caffemodel   -iterations 10

And you can see the benchmark of AlexNet on Pi3 caffe.


Modified 2019-09-22 by Andrea Censi

Step 1: install dependencies and clone repository

First, update apt-get:

$ sudo apt-get update

For Bazel:

$ sudo apt-get install pkg-config zip g++ zlib1g-dev unzip

For Tensorflow: (NCSDK only support python 3+. I didn’t use mvNC on rpi3, so here I choose python 2.7)

(For Python 2.7)

$ sudo apt-get install python-pip python-numpy swig python-dev
$ sudo pip install wheel

(For Python 3.3+)

$ sudo apt-get install python3-pip python3-numpy swig python3-dev
$ sudo pip3 install wheel

To be able to take advantage of certain optimization flags:

$ sudo apt-get install gcc-4.8 g++-4.8
$ sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.8 100
$ sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.8 100

Make a directory that hold all the thing you need

$ mkdir tf

Step 2: build Bazel

Modified 2019-09-22 by Andrea Censi

Download and extract bazel (here I choose 0.7.0):

$ cd ~/tf
$ wget
$ unzip -d bazel

Modify some file:

$ cd bazel
$ sudo chmod u+w ./* -R

$ nano scripts/bootstrap/

To line 117, add “-J-Xmx500M”:

$ nano tools/cpp/cc_configure.bzl

Place the line return “arm” around line 133 (beginning of the _get_cpu_value function):

Build Bazel (it will take a while, about 1 hour):

$ ./

When the build finishes:

$ sudo cp output/bazel /usr/local/bin/bazel

Run bazel check if it’s working:

$ bazel

Step 3: compile Tensorflow

Modified 2019-09-22 by Andrea Censi

Clone tensorflow repo (here I choose 1.4.0):

$ cd ~/tf
$ git clone -b r1.4
$ cd tensorflow

(Incredibly important) Changes references of 64-bit program implementations (which we don’t have access to) to 32-bit implementations.

$ grep -Rl 'lib64' | xargs sed -i 's/lib64/lib/g'

Modify the file platform.h:

$ sed -i "s|#define IS_MOBILE_PLATFORM|//#define IS_MOBILE_PLATFORM|g" tensorflow/core/platform/platform.h

Configure the build: (important) if you want to build for Python 3, specify /usr/bin/python3 for Python’s location and /usr/local/lib/python3.x/dist-packages for the Python library path.

$ ./configure

Build the Tensorflow (this will take a LOOOONG time, about 7 hrs ):

$ bazel build -c opt --copt="-mfpu=neon-vfpv4" --copt="-funsafe-math-optimizations" --copt="-ftree-vectorize" --copt="-fomit-frame-pointer" --local_resources 1024,1.0,1.0 --verbose_failures tensorflow/tools/pip_package:build_pip_package

After finished compiling, install python wheel:

$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
$ sudo pip install /tmp/tensorflow_pkg/tensorflow-1.4.0-cp27-none-linux_armv7l.whl

Check version:

$ python -c 'import tensorflow as tf; print(tf.__version__)'

And you’re done! You deserve a break.

Step 3: try it out

Suppose you already have inception-v3 model (with inception-v3.meta and inception-v3.ckpt)

Create a testing python file

$ vim

Write the following code:

Save, and excute it

$ python cat.jpg

Then it will show the predict label and predict time.