$ node download.js # download images from info.txt obtained from cfw webpage
$ node verify.js # verify all downloaded images, and produce verified.txt index file
** Use face detector to produce directory extracted/
** Manually verify to produce directory annochecked/
$ python remove_spaces.py # ensure all directory names don't contain spaces
$ python split.js # produce train/val splits
$ ./gen_lmdb.sh # (optional) generate caffe lmdb if required + image mean