search

akarshzingade / image-similarity-deep-ranking

369 stars 103 forks source link

About subsample and produce triplet.txt #16

Open longzeyilang opened 6 years ago

longzeyilang commented 6 years ago

Hi, According to the original，I have two question below: 1) the original use muti-scale model, you demo can not use, why? 2) the original use pairwise relevance scores to choose the better positive and negative image sample, in this demo, do not use the mentioned rules. why? Thanks a lot!

akarshzingade commented 6 years ago

Hey Longzeyilang!

1) Do you mean the parallel smaller network with VGG? If yes, I do have implemented it :) 2) The pairwise relevance score is hand crafted. Unfortunately, they haven't open-sourced that information so couldn't use it.

longzeyilang commented 6 years ago

Thank you! I have another problem, the returned value _batchy of _traingenerator function not use in the train. if batch_y is all the same, it does not matter!

akarshzingade commented 6 years ago

Yes, that is correct! I was supposed to change this. But, somehow forgot about it! Thank you for pointing that out :)