Fig 1.
Appearance changes of the 12 AUs coded in DISFA.
This subject was not a part of the DISFA dataset and was included solely for illustrating these AUs as described in FACS [44, 45]. (The subject pictured has provided written informed consent (as outlined in the PLOS consent form) to publish their image alongside the manuscript.).
Table 1.
Description of the 12 AUs coded in DISFA.
Fig 2.
Demonstrating 66 facial keypoints on an in-house subject, akin to those tracked on DISFA subjects.
Out of those, six keypoints (yellow-colored) are used for registration using affine transformation. (The subject pictured has provided written informed consent (as outlined in the PLOS consent form) to publish their image alongside the manuscript.).
Table 2.
Start and end frame number of different target emotion segments.
Fig 3.
AU consistency along the video timeline in DISFA.
Colored bars represent the different emotion segments and the interval represents the Inter-Segment Gap (ISG).
Fig 4.
Expression of a subject at the start, peak consistent frame, and end of an emotion segment when watching the DISFA stimulus.
This subject was not a part of DISFA dataset and included solely for illustrative purposes, highlighting the non-neutral expressions observed within the Inter-Segment Gap (ISG) of some DISFA subjects. (The subject pictured has provided written informed consent (as outlined in the PLOS consent form) to publish their image alongside the manuscript.).
Fig 5.
Cumulative Distribution Function (CDF) of the AU consistency.
For any given bar, the percentage above indicates the proportion of time points (t = 1,2,…4845) where the AU consistency is less than or equal to the value associated with that bar on the x-axis.
Table 3.
Consistency classes based on AU consistency.
Table 4.
Distribution of the four consistency classes present in different emotion segments.
Fig 6.
Regression lines depicting the correlation between AU consistency and different keypoint-based metrics, using their 4845 data points across the video timeline.
Fig 7.
AU consistency and Average t-statistic along the video timeline (y values normalized for both metrics between [0, 1]).
Colored bars represent the emotion segments extended to their subsequent ISGs.
Table 5.
KL-divergence (row-wise averaged) between κAU distribution table and each of the five keypoint-based metrics ().
Table 6.
Distribution of the CRM metrics in the four consistency classes per emotion. Each entry contains values in the order ().