StableDepth: Scene-Consistent and Scale-Invariant Monocular Depth

Kong 4 Hupan Lab https://stabledepth.github.io Figure 1. StableDepth achieves efficient online monocular depth estimation that produces scene-consistent and scale-invariant predictions frame by frame. As shown in the above challenging scenarios, StableDepth generates stable depth maps with precise geometric details and effectively handles dynamic motions, without the need of accessing future frames.