Classification of Artwork Genres Using Image Data.
- Design classification labels. (Perhaps we can use year intervals for now.)
- Load data vectors and true labels into SciKit-Learn (
sklearn), and begin training.
First, clone this repository and cd into it.
$ git clone https://github.com/hughhan1/artwork.git
$ cd artwork
Next, create a virtual environment and install the necessary pip packages.
$ virtualenv env
$ source env/bin/activate
$ pip install -r requirements.txt
Before we do anything further, we need to retrieve our data. This data is available from
MoMA's GitHub repository in a JSON format, available
here.
Save this file as json/artworks.json.
To scrape images from the MoMA collection, run the following command.
$ python moma.py
The -t or --thumbnails option can be provided to obtain thumbnails instead of
larger images, as shown below.
$ python moma.py -t
Regular sized images will be saved into the images/ directory, and thumbnails will be
saved into the thumbnails/ directory.
Images must be stored in a directory named "images", only containing the revant JPEG images needed to be processed (remove any extranoeos files, ie .*).
To create an h5py dataset file, run the following command.
$ python utils.py "dataset_1 datset_2 ..."
where "dataset_1 datset_2 ..." is a list of datasets (i.e. moma, getty, etc) used, separted by spaces