期刊
PATTERN RECOGNITION LETTERS
卷 148, 期 -, 页码 29-35出版社
ELSEVIER
DOI: 10.1016/j.patrec.2021.04.024
关键词
Semantic segmentation; Scene understanding; Transformer; Multi-scale contextual information
资金
- Air Force Office of Scientific Research [FA2386-19-1-4001]
This paper introduces a novel semantic segmentation network TrSeg based on transformer architecture, which can adaptively capture multi-scale information, outperforming other methods.
Recent effort s in semantic segment ation using deep learning frameworks have made notable advances. However, capturing the existence of objects in an image at multiple scales still remains a challenge. In this paper, we address the semantic segmentation task based on transformer architecture. Unlike exist-ing methods that capture multi-scale contextual information through infusing every single-scale piece of information from parallel paths, we propose a novel semantic segmentation network incorporating a transformer (TrSeg) to adaptively capture multi-scale information with the dependencies on original con-textual information. Given the original contextual information as keys and values, the multi-scale con-textual information from the multi-scale pooling module as queries is transformed by the transformer decoder. The experimental results show that TrSeg outperforms the other methods of capturing multi-scale information by large margins. (c) 2021 Elsevier B.V. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据