GCN论文阅读记录

basic model

半监督分类

Semi-Supervised Classification with Graph Convolutional Networks. ICLR 2017. 论文链接
Thomas N. Kipf, Max Welling.
输入：node features X, Gragh G,部分标签Yl
损失函数：

GCN in re-id

对身体部位建图

PH-GCN: Person Re-identification with Part-based
Hierarchical Graph Convolutional Network
论文链接 arxiv 2019.07

上半网络学习apperance feature ，FN-GCN网络学习 structure feature，然后fuse起来作为最后feature。

FN-GCN建立图方式：分成3个scale，每个scale之间feature距离<=2则建立边，不同scale之间有包含关系则建立边，weight正比与feature间距离。