Anticipating human actions by correlating past with the future with Jaccard similarity measures

We propose a framework for early action recognition and anticipation by\ncorrelating past features with the future using three novel similarity measures\ncalled Jaccard vector similarity, Jaccard cross-correlation and Jaccard\nFrobenius inner product over covariances. Using these combinations of novel\nlosses and using our framework, we obtain state-of-the-art results for early\naction recognition in UCF101 and JHMDB datasets by obtaining 91.7 % and 83.5 %\naccuracy respectively for an observation percentage of 20. Similarly, we obtain\nstate-of-the-art results for Epic-Kitchen55 and Breakfast datasets for action\nanticipation by obtaining 20.35 and 41.8 top-1 accuracy respectively.\n