Hi, i am using train_tripletloss.py and i find that in the train function, it does select_triplets, which gives me say 8 triplets, and enqueue 24 image_path into the queue. however, when running sess.run(loss) on those 24 images, it seems less than 8 triplets actually valid, and sometimes 0 triplets remained so that the loss is 0.0 I think it should remain as 8 triplets...what wrong?
cstaff18
commented
5 years ago
Hi, i am using train_tripletloss.py and i find that in the train function, it does select_triplets, which gives me say 8 triplets, and enqueue 24 image_path into the queue. however, when running sess.run(loss) on those 24 images, it seems less than 8 triplets actually valid, and sometimes 0 triplets remained so that the loss is 0.0 I think it should remain as 8 triplets...what wrong?