This is my attempt at writing a garbage classifier (images of garbage) with the techniques from Andrew Ng's courses, using Tensorflow.
I'm also setting up Blender to generate my data because there doesn't seem to exist a large dataset of labelled images of garbage.
Virtualenv:
virtualenv venv
source /venv/bin/activate
Dependencies:
pip install matplotlib
pip install Pillow
CPU-only dependencies:
pip install tensorflow
For GPU, installation instructions are here and here.
Note that for cuDNN, its a .dpkg
file instead of a tarball.
Get CIFAR-10 Dataset here.
I'm trying to use Blender to generate new training data. I'm using .predef
files for editing Blender scripts in an external editor, as described here.
You need to load every script referenced through import
from within Blender by opening the file in the Blender script editor. There's a workaround for this which I'll eventually implement.