摘要
Due to the limited range of depth sensors, the depth estimators for wild images were trained by ordinal data. Without the need of re-train, we proposed an approach to create seamlessly ordinal depth for 360 wild images. The 360 depth was applied to VR display for improving the virtual perception.