Hi,

Thank you very much for your interest in our work!

We used a consistent dataset throughout all the papers. Here is the link to the dataset (https://cvhci.anthropomatik.kit.edu/~datasets-publisher/published_datasets/face_and_body/BBT_Buffy_ACCIO_VGG_face_reps/VGG2.tar.gz) and a bit more details about BBT and BUFFY.

Considered characters: BBT
character_name = {'howard',  'leonard',    'penny',    'raj',    'sheldon'}; % (Tracks --> label)
character_num_labels = {1, 3, 4, 5, 6}; % Only for BBT-0101, I guess.  % (Tracks --> labelid)

Considered characters: BUFFY
character_name = {'xander','buffy','dawn','anya','willow','giles'}; % (Tracks --> label)
character_num_labels = {1, 2, 3, 4, 5, 6}; % Only for BF-0502, I guess. % (Tracks --> labelid)

features_vgg: Feat_Dimension x Num_Frames x Num_Tracks
Num_Frames: Number of frames for a given track can be known via: Tracks -> numframes. Beyond numframes are zero vectors in each track, sorry for the bad structure, created it a long time back 😉 

Please let me know if you need any help.

best, Vivek