The entire dataset, gzipped, can be downloaded at https://goo.gl/VezNNA.
The directory structure correspond to the sources on the Internet from which we grabbed the MIDI files. We deduplicated the dataset and the resulting collection has 77153 songs.
Big_Data_Set: “The Largest MIDI Collection on the Internet”
cariart: Ben Burgraff’s Midi Collection
download-midi: Pop and Rock genres from DOWNLOAD-MIDI.COM
ics: Dan Hirschberg at University of California, Irvine
lmd_matched: Straight from the LMD-matched dataset from the Lakh project
TV_Themes_www.tv-timewarp.co.uk_MIDIRip: (Now defunct) TV Timewarp