RefineNet

Multi-Path Refinement Networks for High-Resolution Semantic Segmentation

RCUMulti-resolution fusionChained residual poolingOutput convolutions
  21

Contributor

contributed at 2020-03-08

Authorship

Affiliation:  
Nanyang Technological University, Singapore.
Email:  
guosheng.lin{At}gmail.com
Homepage:  
View
Is authorship not correct? Feed back

Classification(s)

Geography SubjectGIScience & Remote SensingImagery Processing

Detailed Description

  • 为了解决下采样过程中导致的信息损失,提出了RefineNet,通过利用下采样过程中能够获取到的所有信息。网络组建使用了恒等映射的残余连接。
  • 在公共数据集(VOC 2012)中做了实验,实现了最佳效果。
    • 提出了RefineNet,它是一种多路径的提炼网络,利用多级抽象特征进行高分辨率的语义分割,通过递归方式提炼低分辨率的特征,生成高分辨率的特征。级联的refineNet可以end-to-end训练,使用了恒等映射的残余连接。提出了链式残余池化。使用不同尺寸的窗口池化,并且使用残余连接和可学习的权重把他们融合起来。

References

How to cite

Guosheng Lin (2020). RefineNet, Model Item, OpenGMS, https://geomodeling.njnu.edu.cn/modelItem/68bb26d9-3b79-40e2-9d61-d8a1ceb55cd9
Copy

QR Code

Contributor

contributed at 2020-03-08

Authorship

Affiliation:  
Nanyang Technological University, Singapore.
Email:  
guosheng.lin{At}gmail.com
Homepage:  
View
Is authorship not correct? Feed back

QR Code

You can link related {{typeName}} from your personal space to this model item, or you can create a new {{typeName.toLowerCase()}}.

Model Content & Service

These authorship information will be submitted to the contributor to review.