Question: distance in triplet loss

[mella30]

Hi @lingorX ,
I'm adapting your loss to my framework, and I'm wondering about the distance measure used in the tree triplet loss. In the paper you wrote, that you used the cosine distance but in the code, I only find this snippet:

HieraSeg/Pytorch/mmseg/models/losses/tree_triplet_loss.py

Line 49 in cc3c1cf

distance[:,0:1] = 1-(feats_anchor*feats_pos).sum(1, True)

Which one did you use to produce the reported results? Did you have any insights on the choice of the distance measure?