search

BruceW91 / CVSE

The official source code for the paper Consensus-Aware Visual-Semantic Embedding for Image-Text Matching (ECCV 2020)
172 stars 19 forks source link

About Graph Convolution #5

Closed Liujin21 closed 3 years ago

Liujin21 commented 3 years ago

Hello author, first of all thank you for your open source code. I have a question about the graph convolution part. I saw that in your paper, each layer of graph convolution must be followed by an activation function, but the second layer of the code you gave Graph convolution does not add activation function. I want to know why? Can I add it?

BruceW91 commented 3 years ago

Hi, I followed the Thomas Kipf to use the standard graph convolution in my implementation besides its softmax function. We feel sorry that it's not rigorous to define the graph convolution in our paper (the definition of graph convolution on last layer should be a exception). You can also try to add the activation function on the last layer and I haven't tried it.

Liujin21 commented 3 years ago

Thank you so much for taking the time to reply to me.