Training-by-Fitting: Self-Supervision for 3D Hand Pose Estimation

Add to your list(s) Download to your calendar using vCal

• Dr Angela Yao, National University of Singapore
• Tuesday 24 September 2019, 14:00-15:00
• Room 503, 5th Floor, EEE Building.

If you have a question about this talk, please contact Joan P O'Brien.

Learning based hand pose estimation methods, especially deep learning based methods, requires large amount of accurate annotations on real-world data to achieve high accuracy. However, acquiring such accurate annotated samples can be extremely difficult and expensive. To mitigate the dependency on large amounts of annotation, we propose to leverage unlabeled samples instead.  We propose a method that bridges model-based optimization and discriminative learning by using model-fitting errors to train deep neural networks.  We demonstrate the ability of our method to generalize on two different models, one from a set of spheres and one from triangular meshes. Our proposed method makes highly accurate pose estimates comparable to current supervised methods and advances state-of-the-art in unsupervised learning for hand pose estimation.

Bio Dr Angela Yao has joined the Department of Science at National University of Singapore on 1 October 2018 as an Assistant Professor. She has obtained her PhD and Master of Science degree in Biomedical Engineering from Swiss Federal Institute of Technology (ETH), Zurich in 2012 and 2008, respectively. She obtained her Bachelor’s degree in Engineering Science from the University of Toronto, Canada in 2006. Prior to joining the Dept, Dr Yao was an Assistant Professor at the University of Bonn, Germany. Her research interests include human and hand pose estimation, action recognition, random forests, semi-supervised and unsupervised learning algorithms.

This talk is part of the Featured talks series.

This talk is included in these lists:

• Andrea Picciau's list
• COMMSP & CP list
• COMMSP Seminar
• Featured talks
• Room 503, 5th Floor, EEE Building
• SAP Talks

Note that ex-directory lists are not shown.