Open tkuri opened 3 years ago
RGB+疎なDepthで密なDepth推定をする。従来のように前段でFusionしてE2Eで学習させるのではなく、疎なDepthを最大限活用するために疎Depthを拡張するための領域の信頼性をパッチ毎に推定するアプローチを取る。様々なガイド付きのDepth推定手法に適用できる汎用性がある。
https://openaccess.thecvf.com/content/CVPR2021/html/Huang_S3_Learnable_Sparse_Signal_Superdensity_for_Guided_Depth_Estimation_CVPR_2021_paper.html
未確認。
論文概要
RGB+疎なDepthで密なDepth推定をする。従来のように前段でFusionしてE2Eで学習させるのではなく、疎なDepthを最大限活用するために疎Depthを拡張するための領域の信頼性をパッチ毎に推定するアプローチを取る。様々なガイド付きのDepth推定手法に適用できる汎用性がある。
https://openaccess.thecvf.com/content/CVPR2021/html/Huang_S3_Learnable_Sparse_Signal_Superdensity_for_Guided_Depth_Estimation_CVPR_2021_paper.html
Code
未確認。