Open tkuri opened 2 years ago
単眼Depth推定(SIDE)の性能を上げるために、空間上で互いに近い画像パッチ間の関係性をEdgeConvを用いて学習し構造情報を抽出する。SIDEにEdgeConvを用いたのは初。NYUv2とKITTIデータセットでSOTA。
https://openaccess.thecvf.com/content/WACV2022/html/Lee_EdgeConv_With_Attention_Module_for_Monocular_Depth_Estimation_WACV_2022_paper.html
未確認。
論文概要
単眼Depth推定(SIDE)の性能を上げるために、空間上で互いに近い画像パッチ間の関係性をEdgeConvを用いて学習し構造情報を抽出する。SIDEにEdgeConvを用いたのは初。NYUv2とKITTIデータセットでSOTA。
https://openaccess.thecvf.com/content/WACV2022/html/Lee_EdgeConv_With_Attention_Module_for_Monocular_Depth_Estimation_WACV_2022_paper.html
Code
未確認。