Deep Learning on Small Datasets without Pre-Training using Cosine Loss.

Bjrn Barz, Joachim Denzler

Browse the full WACV paper archive.