YouTube Facial Palsy (YFP) Database
What is Facial Paralysis?
Also known as Bell’s palsy, which is the most common type of facial paralysis, affecting 1/5,000 people a year.
Temporary loss of muscle control in one side of the face, causing asymmetric facial deformation.
Causes are not completely known, but herpes or syphilis could be responsible.
People with diabetes and HIV are at higher risk of developing facial paralysis.
Our Contributions
First deep-learning approach for facial paralysis detection, able to identify affected regions in a still image and the frequency of the syndrome in a video.
Release of the first public database, labeled by clinicians, for facial paralysis study.
Performance Evaluation
As only 21 patients are available, we adopt the leave-one-person-out (LOPO) protocol that takes 20 patients for training and the remaining one for testing in one session, and in the next session the one in testing is replaced by one who was in the previous training, this process is repeated for all 21 patients, and the performance is measured by the average.
In the experiment with CK+ included in the training, we randomly split the CK+ into five subject independent subsets, and run 5-fold cross validation together with the 21 LOPO tests on the YFP dataset.
Performance comparison with and without CK+ expression dataset, and with and without facial edge feature.
Database Content
32 videos of 21 patients from YouTube, and a few patients have multiple videos.
As the shortest facial palsy session lasts for a second, we convert each video into an image sequence with 6FPS;
Manually labeled local palsy regions when the deformation intensity was considered sufficiently high by clinicians.
The palsy regions were labeled by three independent clinicians, and we used the intersection of the independently cropped regions as the ground truth.
We labeled the intensity observed in each palsy region as 0.5 for low or 1.0 for high.
In addition to the syndrome intensity, we also labeled the palsy regions into Classes Eyes or Mouth.
PUBLICATIONS
Please cite the following paper if you make use of the dataset.
GEE-SERN JISON HSU, JIUNN-HORNG KANG, WEN-FONG HUANG
Deep Hierarchical Network With Line Segment Learning for Quantitative Analysis of Facial Palsy
IEEE Access, vol. 7, 4833-4842, Dec. 2018
Contact
Please send an e-mail to the database administrator and cc. to Prof. Gee-Sern (Jison) Hsu to receive the passcode to unlock the zipped database.. Your Email MUST be sent from a valid University account and include the following text:
Subject: Application to download the YFP database
Name: <your first and last name>
Affiliation: <University where you work>
Department: <your department>
Current position: <your job title>
Email: <must be the email at the above mentioned institution>
Postal Address:
Phone number:
I have read and agreed to follow the restrictions specified in the YFP database webpage. This database will only be used for research purposes. I will not make any part of this database available to a third party. I'll not sell any part of this database or make any profit from its use.
<your signature>
In general, a password will take 3~7 workdays to issue. To avoid problems with our spam filter, make sure that your email is sent from an .edu (or similar) address. Failure to follow the instruction may result in no response.
Prof. Hsu's e-mail: jison@mail.ntust.edu.tw
Database administrator: avlabdba@gmail.com