Multi-criteria online Frame-subset Selection for Autonomous Vehicle Videos

•A general technique for selecting labelled and unlabelled frames from driving videos for semantic segmentation.•Proposed and studied a new objective for incorporating pointwise and pairwise criteria for frame subset selection.•Proposed two new criteria, PW-distinctiveness and PW-uncertainty, for subset selection using deep learning models.•Proposed techniques show impressive compression of 100:4 over keyframes, with  ~ 3% reduction in performance.