Some code to run classification operations on the album art database extracted from Angry Metal Guy.
You need Python with Jupyter notebooks and fastai installed. You should probably use a virtualenv of some sort. Follow instructions from these projects to get set up.
The database I extracted from AMG is included. No guarantees are made about the quality of the data.
Run ./make_indexes.py
first. You can edit this file to make adjustments to how categories are
determined.
Then you'll want to start your Jupyter notebook server and probably open brvtality.ipynb
to
train the neural network. most_brvtal.ipynb
extracts some interesting bits from the results.
brvtality-folder.ipynb
lets you do the same thing as the latter script, but across an arbitrary
folder of images rather than the training data set.
black-death.ipynb
works similarly but for telling black metal apart from death metal.
good-bad.ipynb
attempts to do the same for telling good records apart from bad, but it doesn't
work.
No effort has been made to tidy up the code here, and I have no real idea what I'm doing with the neural network libraries. Seems to work, though.