Learning to Segment from a Few Well-Selected Training Images
- Alireza Farhangfar
- Russ Greiner, Dept of Computing Science; PI of AICML
- Csaba Szepesvari, Department of Computing Science; PI of AICML
We address the task of actively learning a segmentation system: given a large number of unsegmented images, and access to an oracle that can segment a given image, decide which images to provide, to quickly produce a segmenter (here, a discriminative random field) that is accurate over this distribution of images. We extend the standard models for active learner to define a system for this task that first selects the image whose expected label will reduce the uncertainty of the other unlabeled images the most, and then after greedily selects, from the pool of unsegmented images, the most informative image. The results of our experiments, over two real-world datasets (segmenting brain tumors within magnetic resonance images; and segmenting the sky in real images) show that training on very few informative images (here, as few as 2) can produce a segmenter that is as good as training on the entire dataset.
Citation
A. Farhangfar, R. Greiner, C. Szepesvari. " Learning to Segment from a Few Well-Selected Training Images". International Conference on Machine Learning (ICML), June 2009.| Keywords: | brain tumor, machine learning, random fields, active learning, BTAP |
| Category: | In Conference |
| Web Links: | Weblink |
BibTeX
@incollection{Farhangfar+al:ICML09,
author = {Alireza Farhangfar and Russ Greiner and Csaba Szepesvari},
title = { Learning to Segment from a Few Well-Selected Training Images},
booktitle = {International Conference on Machine Learning (ICML)},
year = 2009,
}Last Updated: May 22, 2016Submitted by Russ Greiner
