Open tkuri opened 2 years ago
マルチタスク学習(MTL)は一般的に全てのタスクのGTが得られる事が前提だが、1つのデータセットに全てのGTがないのが普通である。そこでそのような不完全なデータセットに対してもMTLを適用するための戦略を提示。具体的にはラベルがないタスクに対しては敵対的学習を用いる。
https://openaccess.thecvf.com/content/WACV2022/html/Wang_Semi-Supervised_Multi-Task_Learning_for_Semantics_and_Depth_WACV_2022_paper.html
未確認。
論文概要
マルチタスク学習(MTL)は一般的に全てのタスクのGTが得られる事が前提だが、1つのデータセットに全てのGTがないのが普通である。そこでそのような不完全なデータセットに対してもMTLを適用するための戦略を提示。具体的にはラベルがないタスクに対しては敵対的学習を用いる。
https://openaccess.thecvf.com/content/WACV2022/html/Wang_Semi-Supervised_Multi-Task_Learning_for_Semantics_and_Depth_WACV_2022_paper.html
Code
未確認。