论文阅读:Scale-aware and High Quality Monocular Depth Estimation Without Real World Labels
CVPR2022主要贡献通过引入虚拟数据集将尺度信息解耦并估计。方法包含两个训练网络:AMDEnetworkΦ:I→d\Phi:I\rightarrowdΦ:I→d,aPoseEstimationNetworkΩ:Ia,Ib→T^a→b\Omega:{I_a,I_b}\rightarrow\hat{T}_{a\rightarrowb}Ω:Ia,Ib→T^a→b;AScaleNetδ:fΦ→s\de