Implement Labeled Faces in the Wild (LFW)

[EronWright]

Implement a reader for LFW as a loose collection of image files in HDFS. The reader should return:

• a row per image
• a label column, based on the directory name
• a features column, based on the image content, as a raveled (flattened) array with row-major (i.e. 'C') layout.

Consider generalizing the reader to support any two-level directory of images.

One must take a dependency on an image reader. The status-quo is the standard ImageIO framework, but I dislike that it initializes AWT and may cause the current window to lose focus. See the related TwelveMonkeys.

Note: I plan to port the implementation of LFW I wrote for the DL4J Spark ML project, seen here.