----------------------------------
Konstantinos Papoutsakis
----------------------------------
Computer Vision & Robotics Lab
Institute of Computer Science (ICS)
Foundation of Research & Technology (FORTH)
Crete, Greece

www.ics.forth.gr/~papoutsa
www.ics.forth.gr/cvrl/evaco
----------------------------------
Last edited: 10 March 2017
----------------------------------

--------------------------------------------------------------------------------------------------------------------------------------------------------
--> seqs_data folder
The 3D (quaternions) skeletal-based data features (42 features) for the 14 sequences of CMU86 dataset were provided by Wen-Sheng Chu (wschu@cmu.edu) in
https://github.com/l2ior/mocap
Please read the README.md file in the folder for more info.
--------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------------------------------------------------------------------------------
--> pairs_data folder
The data (42D feature vector per frame) for the 91 pairs of long sequences generated by combining the original 14 sequences without repetition.
--------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------------------------------------------------------------------------------
-->distance_mats folder: 
You can calculate the distance matrices on your own based on the data in pairs_data folder, using the pair-wise Euclidean distances of frames 
of the paired sequences based on the raw 42D features per frame.
--------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------------------------------------------------------------------------------
-->gt_all folder: 
Each gt_data_86_a_b.mat file contains the ground truth data for both sequences a and b of the pair.
Each .mat file contains two arrays gt_actions_x of Nx4, where N is the number of actions in the respective sequence and columns are: 
[serial_index, start frame, end frame, label_id] 
Ground truth data is based on the segs array in the respective data files of the data folder.
--------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------------------------------------------------------------------------------
-->gt_common folder: 
Each gt_common_81_a_b.mat file contains an array of common actions of format:
[serial_index, start_frame_in_seq_a, end_frame_in_seq_a, start_frame_in_seq_b, end_frame_in_seq_b, label_id]
--------------------------------------------------------------------------------------------------------------------------------------------------------

--------------------------------------------------------------------------------------------------------------------------------------------------------
-->gt_names folder: 
Each gt_names_81_a_b.mat file contains the action labels (names) for both sequences a and b of that pair.
--------------------------------------------------------------------------------------------------------------------------------------------------------