DPN

DPN

2019/11/12

前言：

ResNet 本質上是特徵重用，DenseNet 的特性則是能提取新特徵。DPN 藉著 Higher Order RNN（HORNN） 將兩種網路拼接在一起，同時擁有兩者的特性，也達到更好的效果。

-----

DenseNet

// Review  DPN — Dual Path Networks (Image Classification)

-----

DenseNet

// Review  DPN — Dual Path Networks (Image Classification)

-----

ResNet (Left) DenseNet (Right)

// Review  DPN — Dual Path Networks (Image Classification)

-----

ResNet (Left) DenseNet (Right)

// Review  DPN — Dual Path Networks (Image Classification)

-----

DPN

// Review  DPN — Dual Path Networks (Image Classification)

-----

# DPN

-----

# DPN

-----

References

◎ 論文

# DPN

Chen, Yunpeng, et al. "Dual path networks." Advances in Neural Information Processing Systems. 2017.

https://papers.nips.cc/paper/7033-dual-path-networks.pdf 

-----

◎ 英文參考資料

Review  DPN — Dual Path Networks (Image Classification)
https://towardsdatascience.com/review-dpn-dual-path-networks-image-classification-d0135dce8817

-----

◎ 簡體中文參考資料  

解读Dual Path Networks（DPN，原创） - 知乎
https://zhuanlan.zhihu.com/p/32702293