We provide a two-session based African body dataset with all possible yaw angles and some challenging conditions for soft biometrics.
Body videos were acquired from a total of 564 African subjects (365 males and 209 females) of a mixture of 28 ethnicities within June 21 and November 29, 2018. Body data were collected in two major sessions, and a good number of subjects featured in the second data collection. Subjects naturally looked different in their second sessions as a result of change in clothing, foot wear, make up, hair style or haircut, emotions, etc.
Majority of the videos were acquired in indoor and outdoor environments under challenging conditions, such as, interference with other objects in the scene; acquisition in evening time or in dark backgrounds under inadequate lighting conditions; acquisition under a canopy, in open air and under too bright sunlight.
Distances of subjects from camera are between 1.2 and 3 meters. Subjects were made to turn around a spot (360 degrees) in the video, so that all possible yaw angles can be captured. The total number of frames are 13,428 consisting of 7,124 male and 6,304 female images.
Please note that only specified sample images of the body data, provided by the Research Director of the BVC laboratory, can be shown or displayed in a report or published academic article. No other images of the body data in the BVC body set can be displayed in an academic article. The body samples to be rather used for display are provided in a separate link.
When the BVC Body dataset is used in any form of research, please the following paper should be cited:
O. Iloanusi and C. Mbah, “Gender Estimation from a Hybrid of Face, Upper and Full Body Images at Varying Body Poses,” in 2019 18th IEEE International Conference On Machine Learning And Applications (ICMLA 2019), Boca Raton, Florida, USA, pp. 1944–1949. https://ieeexplore.ieee.org/abstract/document/8999169