Why can embedding be splited into anchor、positive、negative？

[JasonChenhx]

I can't understand the principle of why can embedding be splited into anchor、positive、negative? I know the embedding is from the network, but I want to know the structure of the data set. Thanks.

[VuHuy-cse-9]

Hi, @JasonChenhx , I saw that in file train_tripletloss.py, line 271, Author implemented function select_triplets, which split embeddings into anchor, positive, and negative. From the code, I guest the structure of embeddings are [A_embed1, A_embed2, ..., A_embedN, B_embed1, B_embed2, ...., B_embedM, C_embed1, C_embed2, ... C_embedL].