How are the triplet input images selected in your proposed architecture? Did you use online hard negative mining strategies to select the triplets? "For each positive pair, we produce 10 triplets" can you please explain a little bit what does it mean?
And in your extended work (Beyond triplet loss) how are the quadruplet (4 input images) input images selected in your proposed architecture?
How are the triplet input images selected in your proposed architecture? Did you use online hard negative mining strategies to select the triplets? "For each positive pair, we produce 10 triplets" can you please explain a little bit what does it mean?
And in your extended work (Beyond triplet loss) how are the quadruplet (4 input images) input images selected in your proposed architecture?