Synchronized NAtural and Posed 2D Facial Expressions
Keywords : Head Pose ; Facial Expression ; Landmarks ; Motion Spotting.
To assess the impact of free head movements on expression recognition performance, SNAP-2DFe presents an innovative acquisition system that captures data simultaneously with and without head movement. Moreover, unlike other datasets, SNAP-2DFe provides entirely real data rather than simulated ones.
CONTEXT
DATA
SNaP-2DFe contains 93.240 images from 1,260 videos of 15 subjects. These videos contain image sequences of faces in frontal and non-frontal scenarios. For each subject, 6 head pose variations (static — no head movement, translation-x — along the x-axis, yaw, pitch, roll — up to 60 degrees, and diagonal — from the upper-left corner to the lower-right corner) combined with 7 expressions (neutral, anger, disgust, fear, happiness, sadness, and surprise) were recorded by two synchronized cameras, resulting in a total of 630 constrained recordings (i.e., without head movements) and 630 unconstrained recordings (i.e., with head movements).
SNaP-2DFe also provides annotations of the stages of expression activation (neutral-onset-apex-offset-neutral). 68 facial landmarks (are the same used for the 300-W corpus) have been initially localized. All the frames have then been individually inspected and, when needed, manually re-annotated in order to compensate for landmark localization errors.
ACKNOWLEDGMENTS
This study provides a publicly accessible reference for annotation. We extend our gratitude to IRCICA for granting access to its facilities for data collection.
USAGE
This version of the dataset is strictly reserved to a non-commercial use.
When publishing results on the dataset, please use the following reference:
Benjamin Allaert, José Mennesson, Ioan Marius Bilasco, Chabane Djeraba – Impact of the face registration techniques on facial expressions recognition – Signal Processing: Image Communication, Elsevier, 2017, <10.1016/j.image.2017.11.002>
Bibtex
@article{allaert2018impact,
title={Impact of the face registration techniques on facial expressions recognition},
author={Allaert, Benjamin and Mennesson, Jose and Bilasco, Ioan Marius and Djeraba, Chabane},
journal={Signal Processing: Image Communication},
volume={61},
pages={44–53},
year={2018},
publisher={Elsevier}
}