search

ohhhyeahhh / SiamGAT

Code for the paper "Graph Attention Tracking". (CVPR2021)
https://openaccess.thecvf.com/content/CVPR2021/papers/Guo_Graph_Attention_Tracking_CVPR_2021_paper.pdf
139 stars 19 forks source link

The problem of understanding formula 4 in the paper. #16

Open mr-xiao-zuo opened 3 years ago

mr-xiao-zuo commented 3 years ago

In the article, why is the template vector h_t multiplied by a_i,j, and what is the meaning of the resulting V_i? image

twotwo2 commented 3 years ago

As you know, a_i,j is a correlation score, the more similar the nodes are, the higher the score is. The aggregated feature V_i can implicitly convey whether the node i belongs to the foreground or not.