Position-Focused-Attention-Network
Position-Focused-Attention-Network copied to clipboard
Question for triplet ranking loss
Hi, First of all, thanks for sharing the code publicly.
In the paper, it says your guys are using triplet ranking loss but in the code it is contrastive loss. Looks like they are not the same thing. Triplet loss need 3 input while contrastive loss needs 2. Am I miss-understanding something here? It will be great if you can show me the file/class that does the triplet loss and how does the anchor and its positive/negative input are selected
I'm relatively new to this area so this question may sounds stupid.
Thank you very much!