Thursday, November 14, 2019

DLA

DLA

2019/11/11

DLA 是論文《Deep layer aggregation》的縮寫。作者提出了 Iterative Deep Aggregation(IDA)和 Hierarchical Deep Aggregation(HDA)兩種連接方式,IDA 用在 stage 之間,HDA 用在 block 之間。相對於 FPN、U-Net 的淺層拼接,作者採用了較深的方法,也藉著實驗證明了效果。可說是 ResNet、DenseNet、DPN 一系列結構的一般化。

-----


// Deep Layer Aggregation — Combining Layers in NN Architectures

-----


// Deep layer aggregation. Cvpr2018

-----


# DLA

-----


# DLA

-----


# DLA

-----


# DLA

-----

References

◎ 論文

# DLA
Yu, Fisher, et al. "Deep layer aggregation." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018.
http://openaccess.thecvf.com/content_cvpr_2018/papers/Yu_Deep_Layer_Aggregation_CVPR_2018_paper.pdf

-----

◎ 英文參考資料  

Deep Layer Aggregation — Combining Layers in NN Architectures
https://medium.com/@mikeliao/deep-layer-aggregation-combining-layers-in-nn-architectures-2744d29cab8 

-----

◎ 日文參考資料 

Deep layer aggregation. Cvpr2018
https://www.slideshare.net/ShinichiroMurakami/deep-layer-aggregation-cvpr2018 

CVPR2018 1000本ノック!(なお37本) - ABEJA Tech Blog
https://tech-blog.abeja.asia/entry/cvpr2018-oral

-----

◎ 簡體中文參考資料

Deep Layer Aggregation _ Out of Memory
http://libanghuai.com/2018/12/17/Deep-Layer-Aggregation/

No comments: