CVPR Daily - Tuesday

27 / 30

Linjie Li 27
Linjie Li is currently pursuing her PhD in Computer Science at
Purdue University. Prior to her Ph.D., she obtained a master’s
degree in Electrical Engineering from UCSD. She was working as
a research assistant at GURU lab in UCSD, focusing on machine
learning, computer vision and neural networks.
In the era of the digital age, we are
constantly forming first impressions on
others by browsing each other's
photos online. Although first
impressions seem to be subjective,
psychological studies have shown that
there is often a consensus among
human in how they perceive
attractiveness, trustworthiness, and
dominance in faces. Are deep learning
models, which have successfully
conquered various visual tasks, also
capable of predicting subjective social
impressions of faces? To answer this
question, we systematically examine
40 social features on faces and use
deep learning models to predict
human first impression on faces.
Employing the internal representations
from pretrained neural networks (for
object classification, face identification,
face landmark detection), we build a
ridge regression model on top of the
extracted features and our model can
successfully predict human social
perception whenever human have
consensus. We further visualise the
key features defining different social
attributes to facilitate an
understanding of what makes a faces
salient in a certain social dimension.
Atalla, will be presented tomorrow
(Wednesday) at the Women in
Computer Vision (WicV) Workshop.
Learning to see faces like humans: modeling the social dimensions of faces