Action Bank: A High-Level Representation of Activity in Video

Authors: Ananth Sadanand and Jason J. Corso
Email: sreemana@buffalo.edu, jcorso@buffalo.edu

Group-wise splits for the UCF50 data set.

Since no specific cross-validation scheme is established for the UCF50 data set, we split the data into five sets, observing group boundaries.  On these five sets, we used five-fold cross-validation.  In each fold, we hold one set for testing and train on the other four.

The 5 sets are listed in files set0.txt through set4.txt.

The script ab_ucf50_groups.py on the http://www.cse.buffalo.edu/~jcorso/r/actionbank performs this cross-validation as well.
