The Official Raspberry Pi Projects Book - Projects_Book_v4

Tutorial

raspberrypi.org/magpi The Official Raspberry Pi Projects Book 133

DEEPDREAMS

Above A magical city in a far-off land

Every input neuron is connected to all neurons of
the first hidden layer, but every connection has a
distinct weight. The higher the weight, the more of
the input neuron’s signal reaches the hidden neuron.
If the total signal strength arriving at a hidden neuron
tops its activation threshold, the hidden neuron fires
a signal to all neurons of the next layer which, in
this simplified example, is the output layer. Again,
the signal intensity depends on the weight of the
connection and the incoming signal strength.
The output layer has one neuron for every object to
be classified, so ten neurons are used to identify the
numbers from zero to nine. If the activation threshold
of an output neuron is surpassed by the weighted
incoming connections, the resulting signal strength is
a measure of confidence in the classification.

Machine learning
The weights and thresholds are initialised randomly,
which can cause very bad classification results.
If an output neuron is activated incorrectly (e.g.
if an image from the training set shows a five, but
the output neuron for two produces a strong signal),
its activation threshold and all the weights of its
incoming connections are adjusted. Then, the error is
propagated back proportionally through all connected
neurons lower in the chain, from the highest hidden
layer down to the input layer. This process is repeated
for all of the images in the training data set until the
results given by the network begin to improve.
Instead of accurately detecting objects in images,
by contrast, DeepDream actually changes the input
image to make it more similar to the objects learned
from the training data.
Imagine a cloud as an input image. Some structure
in the cloud might have a vague similarity to the
features DeepDream associates with a ‘dog-like’
object. If this is the case, the input image is changed
to look more dog-like.
Additionally, DeepDream allows for selection of the
hidden layer depth. Since a layer has more detail the
closer it is to the output level, output images can range
from basic shapes to detailed dream-like creatures.

Have some DeepDreams

Insert an SD card with Raspbian installed into

the Raspberry Pi, attach the Camera Module and

peripherals, then boot it up. To install the DeepDream

software, enter the following in a Terminal:

mkdir ~/deepdream && cd ~/deepdream

git clone https://github.com/JoBergs/PsyCam

cd PsyCam

python install_tools.py packages

python install_tools.py caffe

python install_tools.py protobuf

python install_tools.py camera

sudo reboot

After downloading the project from GitHub, use

the custom installer to first install all packages with

pip or apt-get. Then, install the open-source neural

network framework Caffe. Because speed matters, you

should also install the serial data processor protobuf

from Google. Finally, activate the camera and reboot.

All in all, the installation takes a few hours, so

you’ll need to be patient! The installer should also

work for any newish Ubuntu operating system. If you

encounter problems, try using the manual installation

instructions at magpi.cc/2eCSxDt.

Enter the following, and your Pi can start to dream:

cd ~/deepdream/PsyCam

python psycam.py

The network parameters depth (-d), octave (-o),

and type (-t) are randomized. Add a -c to dream

continuously. Pass -i IMAGE.jpg to use an image

as the base for the dream instead of a snapshot. Find

more information on input arguments by checking the

command-line help:

python psycam.py –help

When the Pi finishes dreaming, the dream and the

original photo are stored in the directory /home/pi/

deepdream/PsyCam/dreams with a timestamp. You

can watch them by opening the directory in the file

browser and double-clicking the image.

Language

>PYTHON

DOWNLOAD:

magpi.cc/2eCSQOD

ORIGINAL

DEEPDREAM

ARTICLE

Have a look at

the amazing

original

article about

DeepDream

by Alexander

Mordvintsev,

Christopher

Olah, and Mike

Tyka here:

magpi.cc/

2eCRHH3

NEURAL

NETWORK

TUTORIAL

If you want to

learn more

about neural

networks,

why not

check out this

interesting and

well-written

introduction

by Michael

Nielsen?:

magpi.cc/

2eCUO1u

KIVY GUI

Try the

powerful

Python GUI

framework

Kivy (kivy.org)

for building

a GUI: it is

slick enough

to embrace

DeepDream.

Alice in a very colourful wonderland