eduzhai > Applied Sciences > Engineering >

Multi-Face Self-supervised Multiview Adaptation for Robust Face Clustering in Videos

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 17 pages

Abstract: Robust face clustering is a key step towards computational understanding ofvisual character portrayals in media. Face clustering for long-form contentsuch as movies is challenging because of variations in appearance and lack oflarge-scale labeled video resources. However, local face tracking in videos canbe used to mine samples belonging to same different persons by examining thefaces co-occurring in a video frame. In this work, we use this idea ofself-supervision to harvest large amounts of weakly labeled face tracks inmovies. We propose a nearest-neighbor search in the embedding space to minehard examples from the face tracks followed by domain adaptation usingmultiview shared subspace learning. Our benchmarking on movie datasetsdemonstrate the robustness of multiview adaptation for face verification andclustering. We hope that the large-scale data resources developed in this workcan further advance automatic character labeling in videos.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×