image processing - Training a Siamese Network with Caffe -
i training simple siamese network compares pair of images. followed example given caffe(siamese) , made own model.
my issue constrastive loss function. detail of function implementation caffe defined here. in implementation used margin = 1, defined follows
layer { name: "loss" type: "contrastiveloss" bottom: "data" bottom: "data_p" bottom: "label" top: "loss" contrastive_loss_param { margin: 1 } }
my data labeled 0 if dissimliar , 1 if similar. confused margin of contrastive loss function. how margin parameter selected?
page 3 of initial paper hadsell et.al states margin > 0 there upper bound?
Comments
Post a Comment