search

yuweihao / KERN

Code for Knowledge-Embedded Routing Network for Scene Graph Generation (CVPR 2019)
MIT License
120 stars 35 forks source link

Understanding the equations in the paper. #16

Closed kunalpagarey closed 4 years ago

kunalpagarey commented 4 years ago

@yuweihao I was reading your paper (KERN) and wanted to make sure that there is no mistake in equation 6. You explain it as: all correlated output feature vectors are aggregated to predict the class label but you have also used the hidden state of the last class i.e. h_iC, I am confused. Could you please clarify it.

Thanks.

yuweihao commented 4 years ago

For i-th object region, we use C nodes in the graph to represent the region, namely, h_i1, h_i2, h_i3, ..., h_iC. If we have O object region in the image, the graph for object classification will have C*O nodes. After graph propagation, for each region, we concat all output features of the C nodes of the region for object classification.

kunalpagarey commented 4 years ago

So in this equation e5 there should not be h_iC in the end instead it should be f_iC, right?

yuweihao commented 4 years ago

Yes, you are right. This is a typo. It should be $f_iC$.