
Alignments of the BU-4DFE Database calculated with an algorithm based on:

[1] Awiszus, Graßhof, Kuhnke, and Ostermann. "Unsupervised Features for Facial 
Expression Intensity Estimation over Time". Conference on Computer Vision and
Pattern Recognition Workshop, 2018. CVPRW’18. 



Contents:
#---------------------------------------------------------------------------#

- Alignments_100Frames.mat:

contains a 101x6x100 (Persons x Expressions x Frames) array AlignedFrames,
whose values are the appropriate aligned frames from the original sequence.

- Alignments_100Frames.txt:

see above, but ordered as follows: rows = persons, cols = frames, and
expressions are seperated by two blank rows.


- Alignments_half_30Frames.mat:

contains (similar to above) a 101x6x30 array AlignedFrames, whose values
are the frames of the original sequence corresponding only to a full change
from neutral to an expression. This means that some of these frames step in
reverse order, if the second transition was chosen.

- Alignments_half_30Frames.txt:

see above, but ordered as follows: rows = persons, cols = frames, and
expressions are seperated by two blank rows.


- seqLabels.mat:

contains a 101x6 array, whose values describe the the type of sequence (i,j)
is. the interpretation is as follows:

 0  - OK
 1  - Missing beginning transition
 2  - Missing ending transition
 3  - Multiple changes to the expression
4/5 - Some sort of mistake in the tracked feature points

Most of the time, sequences with mistakes (i.e. 4/5) can't be aligned with
the used algorithm so the frames given in the above matrices from those
sequences are probably not correct. If you want to be completely safe,
the frames with the label 3 might not be correct either depending on the 
application.

#---------------------------------------------------------------------------#
Tl;dr - if in doubt, don't use sequences that have labels 3/4/5 at all
#---------------------------------------------------------------------------#

- seqLabels.txt:

see above, but ordered as follows: rows = persons, cols = expressions.



