Training on multi label output vs just taking labels that have score above some threshold?
25 views
Skip to first unread message
Mike Smith
Jul 22, 2015, 2:54:51 AM7/22/15
to Caffe Users
Say you want to label images with multiple tags. Is it better to train a model so every image has a vector associated to it with each positive label being assigned a "1" else "0", or is it better to just iterate over all the classes and see which ones are above some threshold in terms of confidence?
Reply all
Reply to author
Forward
0 new messages