MediaPipeで生成した顔点群に対するConditional VAEを用いた表情操作

書誌事項

タイトル別名
  • MANIPULATION OF FACIAL EXPRESSIONS USING CONDITIONAL VAE ON FACE POINT CLOUDS GENERATED BY MEDIAPIPE

抄録

A large-scale face point cloud tabular data was created by converting two-dimensional facial images of a straight face and a smiling face into a 3D point cloud using MediaPipe. Two types of Conditional VAE (Normal Conditional VAE and Hyperspherical Conditional VAE) were used to mutually deform the created tabular data between the two facial expressions of a straight face and a smiling face, and a comparative evaluation was performed. The results showed that the method using Hyperspherical Conditional VAE was superior from both the neural network and human perspectives.

収録刊行物

詳細情報 詳細情報について

  • CRID
    1390578050264140544
  • DOI
    10.15002/00026407
  • HANDLE
    10114/00026407
  • ISSN
    24368083
  • 本文言語コード
    ja
  • データソース種別
    • JaLC
    • IRDB
  • 抄録ライセンスフラグ
    使用可

問題の指摘

ページトップへ