We propose a new criterion for identifying the so-called "neglected" classes during the training of DNNs, i.e. the classes which stop to optimize early in the training procedure. Moreover, based on ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results