Adapting to Intra-Class Variations using Incremental Retraining with Exploratory Sampling

Young-Woo Seo, Christopher Urmson, David Wettergreen, and Rahul Sukthankar
tech. report CMU-RI-TR-10-36, Robotics Institute, Carnegie Mellon University, October, 2010


Download
  • Adobe portable document format (pdf) (507KB)
Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

Abstract
Variations in appearance can detrimentally impact the accuracy of object detectors leading to an unacceptably high rate of missed detections. We propose an incremental retraining method that combines a self-training strategy with an uncertainty-based model for active learning. This enables us to augment an existing training set with selectively-labeled instances from a larger pool of examples that exhibit significant intra-class variation while minimizing the user’s labeling effort. Experimental results on an aerial imagery task demonstrate that the proposed method significantly improves over conventional passive learning techniques. Although the experiments presented in this paper are in the domain area of visual object recognition, our method is completely general and is applicable to a broad category of problems in machine learning.

Keywords
Handling of Intra-class Object Appearance Variation, Aerial Image Analysis, Computer Vision, Machine Learning

Notes
Sponsor: GM-CMU AD CRL
Associated Center(s) / Consortia: Field Robotics Center
Associated Project(s): Enhanced Road Network Data from Overhead Imagery

Text Reference
Young-Woo Seo, Christopher Urmson, David Wettergreen, and Rahul Sukthankar, "Adapting to Intra-Class Variations using Incremental Retraining with Exploratory Sampling," tech. report CMU-RI-TR-10-36, Robotics Institute, Carnegie Mellon University, October, 2010

BibTeX Reference
@techreport{Seo_2010_6717,
   author = "Young-Woo Seo and Christopher Urmson and David Wettergreen and Rahul Sukthankar",
   title = "Adapting to Intra-Class Variations using Incremental Retraining with Exploratory Sampling",
   booktitle = "",
   institution = "Robotics Institute",
   month = "October",
   year = "2010",
   number= "CMU-RI-TR-10-36",
   address= "Pittsburgh, PA",
}