ShihaoShao-GH/SuperGlobal

Re-ranking network

Closed this issue · 3 comments

yejy53 commented

Thank you for your great work! I am very interested in your re-ranking network. It seems to be weighted for retrieval features through preliminary ranking results, without involving more complex network optimization processes. Is my understanding correct? Is β among them learnable?

image
R

yejy53 commented

I understand the beta settings! But what I don't know is that gi is implemented by calculating distances between database features?

ShihaoShao-GH commented

Hi. Thanks for liking our work! $g_*$'s across this paper denote the global features of given images. The similarities between g's are determined by the dot product with the vectors g's being of unit length.

yejy53 commented

Thank you for your excellent work. I still don’t understand it. It seems that your reordering network does not require training and has no corresponding loss.