论文阅读笔记:“Spatiotemporal Multiplier Networks for Video Action Recognition”
sixwalter Lv6

Spatiotemporal Multiplier Networks for Video Action Recognition

主要思想

  • to solve the combination of two streams:

multiplicative motion gating(乘法运动门控)

image-20221125091530994

如图2d,其公式为:

image-20221125091658691

在反向传播过程中,运动和表观流的输入被显示地牵涉,作为其梯度上的门控机制。它使得模型有能力去学习时空特征关联

image-20221125093006604
  • to solve long-term input

inject temporal filters(注入时序过滤器)

image-20221125092129508

对应该变换的初始化,作者令其不会改变原始特征(即恒等映射),思想其实就是进行一个3D卷积,只在通道上和时序上进行了信息处理(不涉及空间维度信息变换)。

image-20221125093141403
  • Post title:论文阅读笔记:“Spatiotemporal Multiplier Networks for Video Action Recognition”
  • Post author:sixwalter
  • Create time:2023-08-05 11:14:26
  • Post link:https://coelien.github.io/2023/08/05/paper-reading/paper_reading_057/
  • Copyright Notice:All articles in this blog are licensed under BY-NC-SA unless stating additionally.
 Comments