*******************************************************************************
Instruction for BSS2 (Nov. 5, 2012)
Yagi laboratory, Department of Intelligent Media
The Institute of Scientific and Industrial Research, Osaka University
http://www.am.sanken.osaka-u.ac.jp/index.html
*******************************************************************************

-------------------------------------------------------------------------------
1. Contents
-------------------------------------------------------------------------------
BSS2 is composed of dissimilarity score matrices from 3,706 subjects, which is a full set of OULP-V1C1-A55 defined in [1]. The following dissimilarity score matrices were computed from the individual gait features, which were extracted from the silhouette sequences of each subject.

- BSS2_DissimilarityScoreMatrix_GEI.csv: GEI (gait energy image)

- BSS2_DissimilarityScoreMatrix_FDF.csv: FDF (frequency-domain feature)

- BSS2_DissimilarityScoreMatrix_FDF_Order1.csv: FDF-1 (frequency-domain feature, one-time frequency component)

- BSS2_DissimilarityScoreMatrix_FDF_Order2.csv: FDF-2 (frequency-domain feature, two-time frequency component)

- BSS2_DissimilarityScoreMatrix_GEnI.csv: GEnI (gait entropy image)

- BSS2_DissimilarityScoreMatrix_CGI.csv: CGI (chrono-gait image)

- BSS2_DissimilarityScoreMatrix_GFI.csv: GFI (gait flow image)


Moreover, the subject ID lists for probe and gallery were provided as follows, respectively.

- BSS2_ProbeIDList.csv

- BSS2_GalleryIDList.csv


Detailed procedures of preprocess and feature extraction are found in [1]. 

[1] H. Iwama, M. Okumura, Y. Makihara, and Y. Yagi, ``The OU-ISIR Gait Database Comprising the Large Population Dataset and Performance Evaluation of Gait Recognition,'' IEEE Trans. on Information Forensics and Security, Vol. 7, No. 5, pp. 1511-1521, Oct. 2012.


-------------------------------------------------------------------------------
2. Protocols
-------------------------------------------------------------------------------

-------------------------
2.1 Protocol 1 (BSS2-P1)
-------------------------

Three-dimensional dissimilarity scores from GEI, FDF-1, and FDF-2 are fused in this protocol.

Subject ID lists for training (development) and test (evaluation) of the score-level fusion are provided as follows, respectively.

- BSS2_Protocol1_TrainingIDList.csv

- BSS2_Protocol1_TestIDList.csv

Based on the subject ID lists, sub-matrices of dissimilarity scores for training and test can be extracted for each of GEI, FDF-1, and FDF-2.


-------------------------
2.2 Protocol 2 (BSS2-P2)
-------------------------

Three-dimensional dissimilarity scores from GEI, CGI, and GFI are fused in this protocol.

Subject ID lists for training (development) and test (evaluation) of the score-level fusion are provided as follows, respectively.

- BSS2_Protocol2_TrainingIDList.csv

- BSS2_Protocol2_TestIDList.csv

Based on the subject ID lists, sub-matrices of dissimilarity scores for training and test can be extracted for each of GEI, CGI, and GFI.

-------------------------
2.3 Protocol 3 (BSS2-P3)
-------------------------

Two-dimensional dissimilarity scores from all the combinations of GEI, FDF, GEnI, CGI, GFI are fused in this protocol. More specifically, the following ten combinations are used.

GEI-FDF
GEI-GEnI
GEI-CGI
GEI-GFI
FDF-GEnI
FDF-CGI
FDF-GEF
GEnI-CGI
GEnI-GFI
CGI-GFI

Subject ID lists for training (development) and test (evaluation) of the score-level fusion are provided as follows, respectively.

- BSS2_Protocol3_TrainingIDList.csv

- BSS2_Protocol3_TestIDList.csv

Based on the subject ID lists, sub-matrices of dissimilarity scores for training and test can be extracted for each of GEI, GEnI, FDF, CGI, and GFI.