0%

(CVPR 2018) Squeeze-and-excitation networks

Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7132-7141.



1. Overview


In this paper, it propsed

  • SE block (Squeeze-and-Excitation). adaptively recalibrates channel-wise feature responses by explicity modelling interdependencies between channels
  • SE network



2. Methods




2.1. Squeeze: Global Information Embedding

  • global pooling

2.2. Excitation: Adaptive Recalibration



  • FC + ReLU
  • FC + Sigmoid

2.3. Example




2.4. Implementation





3. Experiments


3.1. Ablation Study



3.2. Comparison