birdortyedi / image-retrieval-with-capsules Goto Github PK

Hi, thanks for sharing the great work!
I figure out that the negative hard sampling is conducted for triplet loss in the paper.
I guess that this part should be implemented in the 'next' method of the 'TripletDirectoryIterator' class. However, what I found in the 'next' method was that negative points are randomly sampled.
Did I find the right code? If not, where could I find the negative hard sampling implementation?

Best regards,

Thanks for your great work, could you please show me your directory structure?
I have trouble in testing because I don't know the test directory

I will cherish your help!