eduzhai > Applied Sciences > Engineering >

History Repeats Itself Human Motion Prediction via Motion Attention

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 19 pages

Abstract: Human motion prediction aims to forecast future human poses given a pastmotion. Whether based on recurrent or feed-forward neural networks, existingmethods fail to model the observation that human motion tends to repeat itself,even for complex sports actions and cooking activities. Here, we introduce anattention-based feed-forward network that explicitly leverages thisobservation. In particular, instead of modeling frame-wise attention via posesimilarity, we propose to extract motion attention to capture the similaritybetween the current motion context and the historical motion sub-sequences.Aggregating the relevant past motions and processing the result with a graphconvolutional network allows us to effectively exploit motion patterns from thelong-term history to predict the future poses. Our experiments on Human3.6M,AMASS and 3DPW evidence the benefits of our approach for both periodical andnon-periodical actions. Thanks to our attention model, it yieldsstate-of-the-art results on all three datasets. Our code is available atthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×