PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Transductive Transfer Learning for Action Recognition
Teo de Campos and Nazli Faraji Davar
In: 25th European Conference on Operational Research, 8-11 Jul 2012, Vilnius, Lithuania.

Abstract

In video processing, it is often the case that one video may contain several samples of actions that share a number of low level features (eg illumination, motion style, etc). Therefore, each video constitutes a domain. We show that, for action classification, a classifier that is trained with one video may lead to poor performance when applied to another video, but if the unlabelled samples of this other video are available, it is possible to apply a transductive transfer learning technique, leading to a significant improvement in performance.

EPrint Type:Conference or Workshop Item (Talk)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Machine Vision
Learning/Statistics & Optimisation
ID Code:9556
Deposited By:Teo de Campos
Deposited On:12 August 2012