论文阅读 [TPAMI-2022] AlignSeg: Feature-Aligned Segmentation Networks

论文阅读 [TPAMI-2022] AlignSeg: Feature-Aligned Segmentation Networks

论文搜索(studyai.com)

搜索论文: AlignSeg: Feature-Aligned Segmentation Networks

搜索论文: http://www.studyai.com/search/whole-site/?q=AlignSeg:+Feature-Aligned+Segmentation+Networks

关键字(Keywords)

Semantics; Context modeling; Computer architecture; Two dimensional displays; Image segmentation; Convolutional codes; Adaptation models; Semantic segmentation; feature alignment; context alignment

机器视觉

检测分割

摘要(Abstract)

Aggregating features in terms of different convolutional blocks or contextual embeddings has been proven to be an effective way to strengthen feature representations for semantic segmentation.

根据不同的卷积块或上下文嵌入聚合特征已被证明是增强语义分割特征表示的有效方法。.

However, most of the current popular network architectures tend to ignore the misalignment issues during the feature aggregation process caused by step-by-step downsampling operations and indiscriminate contextual information fusion.

然而,目前大多数流行的网络架构往往忽略了逐步下采样操作和不加区分的上下文信息融合导致的特征聚合过程中的失准问题。.

In this paper, we explore the principles in addressing such feature misalignment issues and inventively propose Feature-Aligned Segmentation Networks (AlignSeg).

在本文中,我们探讨了解决此类特征错位问题的原则,并创造性地提出了特征对齐分割网络(AlignSeg)。.

AlignSeg consists of two primary modules, i.e., the Aligned Feature Aggregation (AlignFA) module and the Aligned Context Modeling (AlignCM) module.

AlignSeg由两个主要模块组成,即对齐特征聚合(AlignFA)模块和对齐上下文建模(AlignCM)模块。.

First, AlignFA adopts a simple learnable interpolation strategy to learn transformation offsets of pixels, which can effectively relieve the feature misalignment issue caused by multi-resolution feature aggregation.

首先,AlignFA采用一种简单的可学习插值策略来学习像素的变换偏移量,可以有效地缓解多分辨率特征聚合导致的特征错位问题。.

Second, with the contextual embeddings in hand, AlignCM enables each pixel to choose private custom contextual information adaptively, making the contextual embeddings be better aligned.

其次,有了上下文嵌入,AlignCM可以让每个像素自适应地选择私有的自定义上下文信息,从而使上下文嵌入更好地对齐。.

We validate the effectiveness of our AlignSeg network with extensive experiments on Cityscapes and ADE20K, achieving new state-of-the-art mIoU scores of 82.6 and 45.95 percent, respectively.

我们在Cityscapes和ADE20K上进行了大量实验,验证了AlignSeg网络的有效性,分别获得了82.6%和45.95%的最新mIoU分数。.

Our source code is available at https://github.com/speedinghzl/AlignSeg…

我们的源代码可以在https://github.com/speedinghzl/AlignSeg…

作者(Authors)

[‘Zilong Huang’, ‘Yunchao Wei’, ‘Xinggang Wang’, ‘Wenyu Liu’, ‘Thomas S. Huang’, ‘Humphrey Shi’]

你可能感兴趣的:(人工智能,机器学习,深度学习,计算机视觉,CVPR)