You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Forward:
if aps[i][1] >= ans[i][0]:
self.no_residual_list.append(i)
Backward:
for triplet-pair in no_residual_list, gradient is set to zeros
Does "aps[i][1] >= ans[i][0]" mean the distance between anchor and positive sample is larger than the distance between anchor and negative sample? If so, why drop these hard pairs (larger loss pair)?
Another question, u use the sort the positive and negative samples according to the distance between them and anchor, but all these pairs are sent to tripletloss layer, what does this for?
The text was updated successfully, but these errors were encountered:
Forward:
if aps[i][1] >= ans[i][0]:
self.no_residual_list.append(i)
Backward:
for triplet-pair in no_residual_list, gradient is set to zeros
Does "aps[i][1] >= ans[i][0]" mean the distance between anchor and positive sample is larger than the distance between anchor and negative sample? If so, why drop these hard pairs (larger loss pair)?
Another question, u use the sort the positive and negative samples according to the distance between them and anchor, but all these pairs are sent to tripletloss layer, what does this for?
The text was updated successfully, but these errors were encountered: