Abstract
This paper examines the difficulty associated with
performing machine-based automatic demographic prediction
on a sub-population of Asian faces. We introduce the Wild
East Asian Face dataset (WEAFD), a new and unique dataset
to the research community. This dataset consists primarily of
labeled face images of individuals from East Asian countries,
including Vietnam, Burma, Thailand, China, Korea, Japan,
Indonesia, and Malaysia. East Asian turk annotators were
uniquely used to judge the age and fine grain ethnicity attributes
to reduce the impact of the other race effect and improve
quality of annotations. We focus on predicting age, gender and
fine-grained ethnicity of an individual by providing baseline
results with a convolutional neural network (CNN). Finegrained
ethnicity prediction refers to predicting ethnicity of
an individual by country or sub-region (Chinese, Japanese,
Korean, etc.) of the East Asian continent. Performance for two
CNN architectures is presented, highlighting the difficulty of
these tasks and showcasing potential design considerations that
ease network optimization by promoting region based feature
extraction.
