Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation

2024-04-30 13:42:50

A new way to generate CAMs by equivalent after affine

PCM ( Pixel Correlation Module):

less functions:

avoid over-fitting

θ\thetaθ

1 × 1 convolution layers.(?)
x and y denote input and output feature
refine pixel-wise prediction results
y denotes the original CAM and y denotes the revised CAM
What is x?

no residual:

keep the same activation intensity of the original CAM

ReLU:

?We use ReLU activation function with L1 normalization to mask out irrelevant pixels and generate an affinity attention map which is smoother in relevant regions.

ER and ECR Loss

Compare two CAMs whose input is A(img) and img, A is a transform, we need f(A(img))=A(f(img)).

About ECR:
origin one compare with A+PCM.
origin+PCM compare with A.
as PCM fall into the local minimum quickly that all pixels in the image are predicted the same class.

CIs Loss

multi-label soft margin loss (sigmoid + BCEloss)

Finally merge three Loss

Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation相关推荐

阅读笔记：Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation
Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation 基于等变注意力机 ...
Self-supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation论文解读
(CVPR 2020|中科院VIPL实验室) 1.要解决的问题: 基于类别标签的弱监督语义分割是一个具有挑战性的问题,类别响应图(class activation map,简称CAM)始终是这一领域的 ...
【论文阅读】Online Attention Accumulation for Weakly Supervised Semantic Segmentation
一篇弱监督分割领域的论文,其会议版本为: (ICCV2019)Integral Object Mining via Online Attention Accumulation 论文标题: Online ...
TS-CAM: Token Semantic Coupled Attention Map for Weakly Supervised Object Localization
TS-CAM:由Vision Transformer架构产生CAM类别激活图的一种方法文章目录 TS-CAM:由Vision Transformer架构产生CAM类别激活图的一种方法前言一.论文 ...
【2019CVPR学习】翻译-Graph Attention Convolution for Point Cloud Semantic Segmentation
Graph Attention Convolution for Point Cloud Semantic Segmentation-原文链接翻得不好,欢迎批评指正. 摘要标准卷积因为有着特征各向同 ...
[论文阅读] Structure-Consistent Weakly Supervised Salient Object Detection with Local Saliency Coherence
论文地址:https://ojs.aaai.org/index.php/AAAI/article/view/16434 代码:http://github.com/siyueyu/SCWSSOD 发表于 ...
【论文阅读】UntrimmedNets for Weakly Supervised Action Recognition and Detection
Abstract 提出 UntrimmedNet ,从Untrimmed视频的视频级标签中直接学习动作识别和检测模型,分为 classification 和 selection 两个模块,可端到端训练 ...
Ideas For Weakly Supervised Object Localization
Ideas For Weakly Supervised Object Localization 最近开始跟着师兄做弱监督学习和医疗影像,近日阅读了几篇文章,与 Object Localization ...
【论文阅读】Regional Semantic Contrast and Aggregation for Weakly Supervised Semantic Segmentation
论文标题: Regional Semantic Contrast and Aggregation for Weakly Supervised Semantic Segmentation 作者信息: 代 ...

最新文章

热门文章