ACRV, The Australian National University

We present a supervised learning to rank algorithm that effectively orders images by exploiting the structure in im-age sequences. Most often in the supervised learning to rank literature, ranking is approached either by analyz-ing pairs of images or by optimizing a list-wise surro-gate loss function on full sequences. In this work we propose MidRank, which learns from moderately sized sub-sequences instead. These sub-sequences contain use-ful structural ranking information that leads to better learnability during training and better generalization dur-ing testing. By exploiting sub-sequences, the proposed MidRank improves ranking accuracy considerably on an ex-tensive array of image ranking applications and datasets. 1