if you try to download the Hollywood2 dataset from the original source, you may find that some files (classes.txt and ClipSets folders) are missing. If so, you can download them here classes.txt: http://ai.stanford.edu/~quocle/classes.txt ClipSets: http://ai.stanford.edu/~quocle/ClipSets.tar.gz http://ai.stanford.edu/~quocle/ClipSets2.tar.gz