Open tkuri opened 3 years ago
ビデオフレームの補間をする際、Optical Flowの推定をせずにPixelShuffleというアテンションベースの単純な特徴量マップ変換に置き換える提案。大きな動きやオクルージョンも効果的に扱うことができる。
https://ojs.aaai.org//index.php/AAAI/article/view/6693
https://github.com/myungsub/CAIN
フレーム補間において動きを明示的に推定しなくてもSOTAとれる時代。
論文概要
ビデオフレームの補間をする際、Optical Flowの推定をせずにPixelShuffleというアテンションベースの単純な特徴量マップ変換に置き換える提案。大きな動きやオクルージョンも効果的に扱うことができる。
https://ojs.aaai.org//index.php/AAAI/article/view/6693
Code
https://github.com/myungsub/CAIN
感想
フレーム補間において動きを明示的に推定しなくてもSOTAとれる時代。