The metadata displayed for each piece includes (when available) a thumbnail image, title, maker, medium, creation date, and accession date.
This data is drawn directly from WCMA's and is presented in its unaltered form, with the exception of the creation date. An exact date is not known for each piece, and in this case, the date displayed is the midpoint of the possible creation date range. More extensive metadata for each piece can be found by searching for the piece title in WCMA's database directly.
Color classifications are based on dominant image color, found using k-means clustering, with exceptions for images where the dominant hue is in the sepia range. This is to prevent the sepia category from becoming too large to be useful as a classification. To handle these cases, the top three dominant colors are found for each image. If the most dominant color is sepia, the second-dominant is used for classification. If that too is sepia, the third is used. An image is classified as sepia only if all three of the most dominant colors fall into that range. It's still a very large number of images, so don't neglect them! View the special collection of sepia images here, and read more about the color classification process here.
Additionally: thanks and credit to the work of Mike Bostock, Sundar Singh, and Paul Brady, a combination of which formed the starter code for the main color graphs layout, and to Andy Barefoot's Medium post for a tutorial on D3 grid basics.
For more information, read the project overview here,
an informal history of the development process here,
project code here, or dive in to the collection!