Do you ever have a hard time telling things apart? Have you tried looking at thousands of pictures of trees and cars, but you still sometimes get them confused? Now you never have to again, because thanks to Deep Learning your computer can sort through these images for you.
x_or_y is a brain dead simple tool for building classifiers to distinguish between things.
To use x_or_y simply think up a set of things you'd like to distinguish:
$ x_or_y "tree" "car"
... lots and lots of processing...
$ ./tree_OR_car /absolute/path/picture_of_tree.jpeg
... more processing...
RESULTS:
#1: [ 98.31%] tree
#2: [ 1.69%] car
That's all there is to it.
x_or_y simply scrapes Bing images (TODO: Check if this is kosher...) for 1,000 images of each class, and then trains a simple convolutional net on them.
Despite the dichotomous name, you may have as many different classes as you want.
TODO: Document all the issues and bugs. For example, currently makeself cds into the temp dir before running my script, so paths input to the resultant binary have to be absolute. :/
This project has a lot of dependencies. For Ubuntu/Debian users:
# Install the Python dependencies.
pip install tensorflow
pip install tflearn
pip install PIL
pip install numpy
# Install additional dependencies.
apt-get install makeself
apt-get install nodejs
# Finally, install scraper.js's dependencies.
# For those unfamiliar with npm, this must be run from
# the root of the repo, so npm can find package.json.
npm install
If you cannot install PIL via pip and you're on Ubuntu/Debian, then sudo apt-get install python-pil might work for you.
If you run x_to_y and it crashes while "Scraping images" then you might have forgotten to install scraper.js's dependencies by running npm install (installs ~150 MiB of dependencies.).
This entire project is Public Domain -- do whatever you want with it.
Inspired by (and blatantly rips off the network architecture of) the infamous r_u_a_bird.py which in turn takes the network architecture from a tflearn example.