Dense-Gated Blocks, towards layers relationships in densely connected blocks

Joseph Faye

mar. 18 2021

Zoom

ResNet [2], Highway networks [3] to DenseNet [4], adding more inter-layer connections besides the direct connection in adjacent layers, emerged as popular approaches to strengthen feature propagation among different layers. However, dense connections cause much redundancy especially in the case of DenseNet. Another aspect is that for many dense connections from previous layers, the role played by the mainstream module is unclear. To address these issues, authors introduce a gating mechanism, inspired by SENeT [5] to model the layer relationships in densely connected blocks.

illustration

[1] JJi, Zhongyi and Han, Xiao and Lin, Tong and Wang, Wenmin. 2020. A Dense-Gated U-Net for Brain Lesion Segmentation.

[2] He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian. 2016. Deep Residual Learning for Image Recognition.

[3] Rupesh Kumar Srivastava and Klaus Greff and Jürgen Schmidhuber. 2015. Highway Networks.

[4] Gao Huang and Zhuang Liu and Laurens van der Maaten and Kilian Q. Weinberger. 2018. Densely Connected Convolutional Networks.

[5] Jie Hu and Li Shen and Samuel Albanie and Gang Sun and Enhua Wu. 2019. Squeeze-and-Excitation Networks.