Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

DrsNet: Dual-resolution semantic segmentation with rare class-oriented superpixel prior

DrsNet: Dual-resolution semantic segmentation with rare class-oriented superpixel prior Rare-class objects in natural scene images that are usually small and less frequent often convey more important information for scene understanding than the common ones. However, they are often overlooked in scene labeling studies due to two main reasons, low occurrence frequency and limited spatial coverage. Many methods have been proposed to enhance overall semantic labeling performance, but only a few consider rare-class objects. In this work, we present a deep semantic labeling framework with special consideration of rare classes via three techniques. First, a novel dual-resolution coarse-to-fine superpixel representation is developed, where fine and coarse superpixels are applied to rare classes and background areas respectively. This unique dual representation allows seamless incorporation of shape features into integrated global and local convolutional neural network (CNN) models. Second, shape information is directly involved during the CNN feature learning for both frequent and rare classes from the re-balanced training data, and also explicitly involved in data inference. Third, the proposed framework incorporates both shape information and the CNN architecture into semantic labeling through a fusion of probabilistic multi-class likelihood. Experimental results demonstrate competitive semantic labeling performance on two standard datasets both qualitatively and quantitatively, especially for rare-class objects. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Multimedia Tools and Applications Springer Journals

DrsNet: Dual-resolution semantic segmentation with rare class-oriented superpixel prior

Loading next page...
 
/lp/springer-journals/drsnet-dual-resolution-semantic-segmentation-with-rare-class-oriented-GmXbm2Jm13

References (71)

Publisher
Springer Journals
Copyright
Copyright © Springer Science+Business Media, LLC, part of Springer Nature 2020
ISSN
1380-7501
eISSN
1573-7721
DOI
10.1007/s11042-020-09691-y
Publisher site
See Article on Publisher Site

Abstract

Rare-class objects in natural scene images that are usually small and less frequent often convey more important information for scene understanding than the common ones. However, they are often overlooked in scene labeling studies due to two main reasons, low occurrence frequency and limited spatial coverage. Many methods have been proposed to enhance overall semantic labeling performance, but only a few consider rare-class objects. In this work, we present a deep semantic labeling framework with special consideration of rare classes via three techniques. First, a novel dual-resolution coarse-to-fine superpixel representation is developed, where fine and coarse superpixels are applied to rare classes and background areas respectively. This unique dual representation allows seamless incorporation of shape features into integrated global and local convolutional neural network (CNN) models. Second, shape information is directly involved during the CNN feature learning for both frequent and rare classes from the re-balanced training data, and also explicitly involved in data inference. Third, the proposed framework incorporates both shape information and the CNN architecture into semantic labeling through a fusion of probabilistic multi-class likelihood. Experimental results demonstrate competitive semantic labeling performance on two standard datasets both qualitatively and quantitatively, especially for rare-class objects.

Journal

Multimedia Tools and ApplicationsSpringer Journals

Published: Sep 9, 2020

There are no references for this article.